# Aaron Makelky full site summary # I teach complex tech to people who don't write code Source: https://aaronmakelky.com Last updated: 2026-09-04 Aaron Makelky helps people and organizations use technology on work they already have. He teaches technical topics to people who do not write code, through content and video. Through workshops, talks, and video. [Book a 30-minute call](https://cal.com/aaron-makelky/intro) [Bring a job](#work-with-me) ![Aaron Makelky at OpenAI Headquarters](/images/blog/openai-codex-party-middle-ai-people/openai-gpt55-stage-aaron.jpg) Aaron Makelky At OpenAI Headquarters Casper, Wyoming Start with real work ## Bring a job your team already does. [Book a 30-minute call](https://cal.com/aaron-makelky/intro) ### Workshops and school training Try your own work in the room. Check the result before anyone calls it useful. ### Speaking I show the work, including the tools I would not use again. ### One job at a time Pick one repeat job. I build a test around how your team actually does it. > Practical usage and protections for AI abuse in the classroom. Super fun, like a conversation with a coworker. Taylor M. Education leader · TeacherCon 25 What I do now ## Complex topics, taught in plain language. I help people and organizations use technology on work they already have. Most of that teaching is content and video. [More about me](/about) ## Astra Commons. Casper. Free coffee, food, and GPT-6 credits. September 12 · 10 a.m. to noon CitiWerks Cafe and Bistro [Register directly on Luma](https://luma.com/event/evt-9gAr2WhAoJOJlid) From my own desk ## Downloads for jobs I got tired of doing by hand. ### Small tools The ones that saved enough time to keep. [Open tools](/tools) ### Field notes Write-ups on tools I tried, including the ones I would not use again. [Read the blog](/blog) ## Related pages - [About Aaron Makelky](/about) - [AI notes](/blog) - [Tools and downloads](/tools) - [OpenClaw kits](/shop) ## Common questions ### Who is Aaron Makelky? Aaron Makelky teaches technical topics to people who don't write code, through content and video. ### What does Aaron help people do with technology? He helps people and organizations use technology on work they already have. --- # Aaron Makelky Source: https://aaronmakelky.com/wiki Last updated: 2026-08-27 Aaron Makelky is an educator, football coach, and workshop facilitator based in Wyoming in the United States. A former Wyoming teacher and football coach, he runs AI workshops, school trainings, speaking, and consulting for people who do not write code. [Wiki The encyclopedia](/wiki) Search Contents 1. [1 Teaching and coaching](#Teaching_and_coaching) 2. [2 Workshops and consulting](#Workshops_and_consulting) 1. [2.1 Approach](#Approach) 3. [3 Public tools](#Public_tools) 4. [4 Writing](#Writing) 5. [5 Software projects](#Software_projects) 6. [6 External links](#External_links) 7. [7 References](#References) - Article - Talk - Read - Edit - View history This page is not a Wikipedia article and is not a Wikimedia project. It is a standalone encyclopedia-style page on aaronmakelky.com. ## Aaron Makelky ![Aaron Makelky](/assets/aaron-headshot-320-DQQOODZH.jpg) Aaron Makelky | Occupations | Educator / Football coach / Workshop facilitator | | --- | --- | | Residence | Wyoming, U.S. | | Website | [aaronmakelky.com](https://aaronmakelky.com) | Aaron Makelky is an educator, football coach, and workshop facilitator based in Wyoming in the United States. [[1]](#cite_note-1) A former Wyoming teacher and football coach, he runs AI workshops, school trainings, speaking, and consulting for people who do not write code. [[1]](#cite_note-1) He works remotely from Wyoming. [[2]](#cite_note-2) His public website describes workshops built around work a group already has: participants try that work in the room and check the result. [[1]](#cite_note-1) He publishes field notes, small tools, and public GitHub repositories. [[3]](#cite_note-3) [[4]](#cite_note-4) ## Teaching and coaching Makelky taught and coached football in Wyoming before working in technology. [[2]](#cite_note-2) His website identifies him as a former Wyoming teacher and football coach. [[1]](#cite_note-1) A public coaching site, 307 Pediatrics Grizzlies Practice Plans, publishes practice plans in the context of the Casper Junior Football League (CJFL). [[5]](#cite_note-5) ## Workshops and consulting The public site lists three ways to work with him: workshops and school training; speaking; and one-job-at-a-time consulting-style tests. [[1]](#cite_note-1) He tests AI tools on real work and turns useful parts into workshops, notes, and small tools. [[2]](#cite_note-2) The about page says testing must handle a rushed calendar, a messy spreadsheet, and a person who does not care how clever a model is. [[2]](#cite_note-2) Intro booking is available via Cal.com. [[6]](#cite_note-6) ### Approach Useful beats impressive; teach the skill not the wrapper; no brand deals and no sponsored praise. [[2]](#cite_note-2) ## Public tools The tools page is titled "Tools I made after the normal way got annoying". [[3]](#cite_note-3) Public tools include: - [Personal API](https://aaronmakelky.com/tools/personal-api) (public JSON, also available at /api/personal) - [Codex Pet Exchange](/tools/codex-pets) - [Mic Pro Wallpaper Maker](/tools/insta360-mic-pro-logo-kit) (Insta360 Mic Pro, 240 x 208) - [X Article Formatter](/tools/x-article-formatter) - [AI workflow checklist](/checklist) - [Presidential Briefing Infographic](/tools/presidential-briefing-infographic) ## Writing Dated titles as published on the blog: [[7]](#cite_note-7) - 2026-07-25. [I edited a fishing video from the ChatGPT mobile app without touching a timeline](/blog/codex-hyperframes-video-editing-workflow) - 2026-04-07. [I cleaned up ten skill folders so you can see the wiring](/blog/10-openclaw-skills-running-on-raspberry-pi) - 2026-03-27. [He told me he felt like he could finally read](/blog/ai-in-education-friction-question-not-tool-question) - 2026-02-20. [revision history may not catch the agent](/blog/agentic-ai-classrooms-what-teachers-need-to-know) - 2026-02-20. [cheaper tools did not make anyone sound like themselves](/blog/podcasting-for-creators-human-content-wins-2026) - 2025-03-03. [most teacher AI apps are wrappers](/blog/3_ai_apps_teachers) - 2025-02-22. [context, a revision, then check the work](/blog/ai_skills_everyone_2025) - 2025-02-17. [AI policy for schools without pretending it is simple](/blog/ai_policy_for_schools) - 2025-02-02. [I made these AI mistakes as a teacher first](/blog/5-mistakes-teachers-make-with-ai) ## Software projects Makelky's GitHub account is a-makelky. As of 27 August 2026 it listed 29 public repositories and the location WYO. [[4]](#cite_note-4) - [ai-writing-audit](https://github.com/a-makelky/ai-writing-audit) - [back-to-school-night-25](https://github.com/a-makelky/back-to-school-night-25) (2025 world history syllabus) - [Steel-Speed-and-Strategy-Quiz](https://github.com/a-makelky/Steel-Speed-and-Strategy-Quiz) (WWII lesson plan with National Museum of Military Vehicles) - [wynpn-skills-2026](https://github.com/a-makelky/wynpn-skills-2026) - [wassp](https://github.com/a-makelky/wassp) - [wyoming-ai-policy-landing-page](https://github.com/a-makelky/wyoming-ai-policy-landing-page) - [agent-huddle](https://github.com/a-makelky/agent-huddle) - [GPT-Realtime-Slides](https://github.com/a-makelky/GPT-Realtime-Slides) [gpt-realtime-slides.pages.dev](https://gpt-realtime-slides.pages.dev/) - [write-better-x-articles](https://github.com/a-makelky/write-better-x-articles) - [notepad-essay](https://github.com/a-makelky/notepad-essay) - [bingo-card-it](https://github.com/a-makelky/bingo-card-it) - [codex-pet-exchange](https://github.com/a-makelky/codex-pet-exchange) - [insta360-mic-pro-logos](https://github.com/a-makelky/insta360-mic-pro-logos) - Public Descript-related repositories: [descript-marketing-brand](https://github.com/a-makelky/descript-marketing-brand), [poke-descript-connector](https://github.com/a-makelky/poke-descript-connector), [descript-streamdeck](https://github.com/a-makelky/descript-streamdeck). ## External links - [Website](https://aaronmakelky.com) - [About](/about) - [Tools](/tools) - [Blog](/blog) - [Links](/links) - [Cal.com intro](https://cal.com/aaron-makelky/intro) - [Personal API JSON](https://aaronmakelky.com/api/personal) - [GitHub](https://github.com/a-makelky) - [X @theaaron](https://x.com/theaaron) - [YouTube](https://www.youtube.com/channel/UC8vjf5_1lXrXYJfD4dICGRg) - [TikTok @aaronmakelky](https://www.tiktok.com/@aaronmakelky) - [Instagram @aaronmakelky](https://www.instagram.com/aaronmakelky) - [LinkedIn](https://www.linkedin.com/in/aaron-makelky-m-a-ed-038b852a3/) - [Cursor](https://cursor.com/@the-aaron) - [Facebook](https://www.facebook.com/profile.php?id=61556718986067) - [practice.aaronmakelky.com](https://practice.aaronmakelky.com) - [creatortoolsdirectory.com](https://creatortoolsdirectory.com) ## References 1. [Aaron Makelky](https://aaronmakelky.com). aaronmakelky.com. Retrieved 27 August 2026. 2. [About Aaron Makelky](https://aaronmakelky.com/about). aaronmakelky.com. Retrieved 27 August 2026. 3. [Tools I made after the normal way got annoying](https://aaronmakelky.com/tools). aaronmakelky.com. Retrieved 27 August 2026. 4. [a-makelky](https://github.com/a-makelky). GitHub. Retrieved 27 August 2026. 5. [307 Pediatrics Grizzlies Practice Plans](https://practice.aaronmakelky.com). practice.aaronmakelky.com. Retrieved 27 August 2026. 6. [Book an intro](https://cal.com/aaron-makelky/intro). Cal.com. Retrieved 27 August 2026. 7. [AI notes](https://aaronmakelky.com/blog). aaronmakelky.com. Retrieved 27 August 2026. 8. [Links](https://aaronmakelky.com/links). aaronmakelky.com. Retrieved 27 August 2026. 9. [Personal API JSON](https://aaronmakelky.com/api/personal). aaronmakelky.com. Retrieved 27 August 2026. 10. [@theaaron](https://x.com/theaaron). X. Retrieved 27 August 2026. 11. [Aaron Makelky](https://www.youtube.com/channel/UC8vjf5_1lXrXYJfD4dICGRg). YouTube. Retrieved 27 August 2026. 12. [@aaronmakelky](https://www.tiktok.com/@aaronmakelky). TikTok. Retrieved 27 August 2026. 13. [@aaronmakelky](https://www.instagram.com/aaronmakelky). Instagram. Retrieved 27 August 2026. 14. [Aaron Makelky](https://www.linkedin.com/in/aaron-makelky-m-a-ed-038b852a3/). LinkedIn. Retrieved 27 August 2026. 15. [the-aaron](https://cursor.com/@the-aaron). Cursor. Retrieved 27 August 2026. 16. [Aaron Makelky](https://www.facebook.com/profile.php?id=61556718986067). Facebook. Retrieved 27 August 2026. 17. [Creator Tools Directory](https://creatortoolsdirectory.com). creatortoolsdirectory.com. Retrieved 27 August 2026. Categories: - Living people - People from Wyoming - Educators - American football coaches - Workshop facilitators [About](/about) [Contact via the website](/links) [Book an intro](https://cal.com/aaron-makelky/intro) [aaronmakelky.com](https://aaronmakelky.com) ## Related pages - [About Aaron Makelky](/about) - [Tools](/tools) - [AI notes](/blog) - [Links](/links) ## Sources and references - [Book an intro](https://cal.com/aaron-makelky/intro) - [GitHub a-makelky](https://github.com/a-makelky) --- # Wyoming teacher to tech Source: https://aaronmakelky.com/about Last updated: 2026-05-28 About page for Aaron Makelky, a former Wyoming teacher and football coach who tests AI tools on real work and teaches the parts that hold up. About I taught and coached in Wyoming before tech. Now I test AI tools on real work, then turn the useful parts into workshops, notes, and small tools. ![Aaron Makelky](/assets/aaron-headshot-320-DQQOODZH.jpg) Aaron Makelky The route ## I was a teacher and coach before I worked in tech. That still explains how I look at AI. If a tool only works when the room is quiet and the prompt is perfect, it is not ready yet. > I work remotely from Wyoming now. That still feels a little funny and a lot useful. Filter 1 ### Test it on a normal Tuesday. A tool has to handle a rushed calendar, a messy spreadsheet, and a person who does not care how clever the model is. A clean demo is not enough. Filter 2 ### Teach the move, not the tour. Most people do not need another walkthrough. They need the first thing to try, what to check, and when to stop. Filter 3 ### Keep the incentives clean. No brand deals. No sponsored praise. If I talk about a tool, it is because I used it and the lesson is worth sharing. How I work ## The filter I keep coming back to. ### Useful beats impressive. If a tool sounds futuristic but does not help someone finish the job or make a better call, it is noise. ### Teach the skill, not the wrapper. I care less about the platform and more about whether someone can use the move again. What I do now ## I test tools and teach the parts that hold up. Most of my work is practical: public tests, workshops, creator workflows, and small resources for people who do not want jargon or product docs. I also do consulting, school trainings, and speaking. Same teaching job as before: make the next step clear enough that somebody can try it after the session ends. Let's talk ## Bring me in when the room agrees AI matters and nobody knows what to try Monday. I can help with workshops, school trainings, creator workflows, or turning one messy process into a next step people can actually run. [Get in touch](mailto:hello@aaronmakelky.com) [Other places I post](/links) ## Related pages - [I teach complex tech to people who don't write code](/) - [AI in education note](/blog/ai-in-education-friction-question-not-tool-question) - [Contact and social links](/links) - [AI tools Aaron publishes](/tools) ## Sources and references - [Descript](https://www.descript.com/) - [Skilling Future Podcast conversation](https://www.youtube.com/watch?v=JoZXbV-w8Bo) ## Common questions ### What is Aaron Makelky known for? Aaron is known for plain-English AI education and practical workflow testing, shaped by his years as a Wyoming teacher and coach before tech. ### Does Aaron do workshops or consulting? Yes. Aaron runs AI workshops, school training, and speaking sessions. He also helps teams test one stuck workflow. --- # Start here if you want to reach me Source: https://aaronmakelky.com/links Last updated: 2026-05-26 Link hub for booking Aaron Makelky's $999 1 Hour AI hour, email, shop, and public profiles. Links Your team keeps saying they should use AI. Nobody knows what that means yet. Book the hour if you want to leave with something built. Email if you just need to write. ![Aaron Makelky](/assets/aaron-headshot-320-DQQOODZH.jpg) 1 Hour AI · $999 ## You've watched the demos. You still don't have a system. We pick one problem and build it together. Live hour on Google Meet. Tools, agents, or automations. [Book this hour](https://cal.com/aaron-makelky/1-hour-ai) Start here ## The easiest ways to reach me. Email ### Send me an email For speaking, consulting, or a note that does not need a calendar invite. [Open](mailto:aaron@aaronmakelky.com) Shop ### Open the shop OpenClaw downloads and checklists. [Open](/shop) X Money ## If this was useful, you can tip me. Scan to send a tip on X Money. X Money ![X Money QR code for tipping Aaron Makelky.](/assets/cash-app-tip-qr-BKUtwFQ1.png) Social links ## Use the channel you already check. ### LinkedIn Aaron Makelky [Open LinkedIn](https://www.linkedin.com/in/aaron-makelky-m-a-ed-038b852a3/) ### YouTube @AaronMakelky [Open YouTube](https://www.youtube.com/channel/UC8vjf5_1lXrXYJfD4dICGRg) ### TikTok @aaronmakelky [Open TikTok](https://www.tiktok.com/@aaronmakelky) ### X @theaaron [Open X](https://x.com/theaaron) ### Instagram @aaronmakelky [Open Instagram](https://www.instagram.com/aaronmakelky) ### Facebook Aaron Makelky [Open Facebook](https://www.facebook.com/profile.php?id=61556718986067) ### Cursor @the-aaron Public profile where I share what I am building in Cursor. [Open Cursor](https://cursor.com/@the-aaron) ## Related pages - [About Aaron](/about) - [Read notes](/blog) - [Shop](/shop) ## Sources and references - [Book 1 Hour AI](https://cal.com/aaron-makelky/1-hour-ai) - [Email Aaron](mailto:aaron@aaronmakelky.com) - [Aaron on LinkedIn](https://www.linkedin.com/in/aaron-makelky-m-a-ed-038b852a3/) - [Aaron on X](https://x.com/theaaron) - [Aaron on Cursor](https://cursor.com/@the-aaron) ## Common questions ### What is the fastest way to contact Aaron? Book the $999 1 Hour AI hour on the links page, or email Aaron directly. ### What is 1 Hour AI? A $999 live hour on Google Meet. Aaron picks one problem with you and builds something useful with AI tools, agents, or automations. ### Can I tip Aaron? Yes. The links page has an X Money QR if the work has been useful. Scan it to send a tip. --- # Notes from AI work I want to find again Source: https://aaronmakelky.com/blog Last updated: 2026-05-26 Blog index for notes Aaron wants to find again: classroom AI, tool tests, agent jobs, creator work, and workday fixes. Blog Tool tests and classroom AI notes, plus the workday fixes I kept needing to look up again. ## All posts September 8, 2026 · 10 min read ### [I paid $300 for Grok Bot. Its cloud computer is the most useful part.](/blog/grok-bot-worth-it-300-dollar-review) I use Grok Bot every day. Here's how its cloud computer helps with my work, where usage limits get in the way, and why I keep my instructions outside the app. ![Aaron Makelky and Ryan Doser discuss Grok Bot: He paid $300 so you don't](/images/blog/grok-bot-review/hero.jpg) AI Agents · Grok Bot [Read](/blog/grok-bot-worth-it-300-dollar-review) September 4, 2026 · 2 min read ### [I'm hosting.](/blog/im-hosting) Astra Commons is coming to Casper. Join me and OpenAIDevs on September 12 at CitiWerks for free coffee, food, and GPT-6 credits. Events · Casper · GPT-6 [Read](/blog/im-hosting) September 3, 2026 · 3 min read ### [How to get more Grok Bot usage (without paying more)](/blog/how-to-get-more-grok-bot-usage) Grok Bot is awesome, but the rate limits suck. Here's a little trick that's totally legit: run Cursor, Grok Build, Codex, or Google's anti-gravity on Grok Bot's own Linux terminal so the coding work does not eat your usage. Grok Bot · Rate limits · Herdr · Agents [Read](/blog/how-to-get-more-grok-bot-usage) August 31, 2026 · 6 min read ### [Your Grok Bot setup is a mess](/blog/how-i-set-up-grok-bot) I organize Grok Bot like an org system with a chief of staff, project leads, and shared specialists. Grok Bot · Workflow · Agents [Read](/blog/how-i-set-up-grok-bot) July 25, 2026 · 12 min read ### [I edited a fishing video from the ChatGPT mobile app without touching a timeline](/blog/codex-hyperframes-video-editing-workflow) The real prompts, screenshots, corrections, and tools behind a 31.7-second vertical video I edited with Codex and HyperFrames from my phone. ![Six frames from Aaron Makelky's Codex and HyperFrames fly-fishing video.](/og/codex-hyperframes-video-editing-workflow.png) Codex · HyperFrames · AI Video Editing · ChatGPT · Creator Workflows [Read](/blog/codex-hyperframes-video-editing-workflow) June 18, 2026 · Transcript ### [Aaron with Kevin, June 18, 2026](/tko-2026-6-18) A timestamped transcript with Aaron and Kevin labelled in the script. Conversation · Transcript [Read](/tko-2026-6-18) June 16, 2026 · 4 min read ### [one card run beat a year of errands](/what-i-built-today-2026-06-16) A night-before anniversary card run left me with a Poke recipe, one dollar-store trip, and a card bench for the next year. What I built today · Poke · Family logistics · Field Notes [Read](/what-i-built-today-2026-06-16) June 15, 2026 · 4 min read ### [the row shape was the build](/what-i-built-today-2026-06-15) I cleaned up a tiny content ops problem that was getting just annoying enough to matter: one promo slot had turned into a pile of platform rows. What I built today · Operations · Content Workflow · Field Notes [Read](/what-i-built-today-2026-06-15) June 4, 2026 · 4 min read ### [the chart needed a memory](/what-i-built-today-2026-06-04) A monthly slide update became a source note: chart, numbers, read of the month, and the manual paste step. What I built today · Analytics · Operations [Read](/what-i-built-today-2026-06-04) June 2, 2026 · 4 min read ### [the comment had no instructions](/what-i-built-today-2026-06-02) A vague assigned comment had no instructions, so I built slide-ready notes. The deck edit stayed manual because that was the right handoff. What I built today · Automation · Analytics [Read](/what-i-built-today-2026-06-02) May 29, 2026 · 4 min read ### [I did not have to ask](/what-i-built-today-2026-05-29) The board already knew what could move and what needed to wait, so the day started without another tab hunt. What I built today · Automation · Planning [Read](/what-i-built-today-2026-05-29) May 28, 2026 · 4 min read ### [no specialists in the room](/what-i-built-today-2026-05-28) A motion piece in Jitter while the designer was out, a Descript API transcript pull to answer a livestream viewer, and a full workshop follow-up that never left Notion. What I built today · Jitter · Descript API · Notion [Read](/what-i-built-today-2026-05-28) May 27, 2026 · 4 min read ### [the dashboard asked first](/what-i-built-today-2026-05-27) Opening the day to a ready-made operating card that pulled the real priorities into one place. What I built today · AI Workflows · Operations [Read](/what-i-built-today-2026-05-27) May 26, 2026 · 4 min read ### [the system stopped living on one laptop](/what-i-built-today-2026-05-26) A chopped-up day left me with a workshop board and a repo the agents could all read. The content plan got smaller too. What I built today · Automation · Planning [Read](/what-i-built-today-2026-05-26) May 25, 2026 · 4 min read ### [the brief was the build](/what-i-built-today-2026-05-25) A small prep automation gave the fuzzy workshop lane a page that was ready on Monday. What I built today · Planning · Field Notes [Read](/what-i-built-today-2026-05-25) May 22, 2026 · 5 min read ### [three tiny systems that made the day less messy](/what-i-built-today-2026-05-22) Four small jobs landed: a watcher, a connector pass, an agent-planning lane, and one hoodie reminder. What I built today · Agents · Automation [Read](/what-i-built-today-2026-05-22) April 7, 2026 · 8 min read ### [I cleaned up ten skill folders so you can see the wiring](/blog/10-openclaw-skills-running-on-raspberry-pi) Hardware is a Pi 5. Each folder has a SKILL.md, a SETUP.md, and scripts only where a prompt is not enough. OpenClaw · Raspberry Pi · Agents [Read](/blog/10-openclaw-skills-running-on-raspberry-pi) March 27, 2026 · 10 min read ### [He told me he felt like he could finally read](/blog/ai-in-education-friction-question-not-tool-question) A student told me the tool let him read. The hard part was deciding which friction was still the work. From a Skilling Future Podcast conversation about AI inside a school. Classroom AI · Literacy · Podcast [Read](/blog/ai-in-education-friction-question-not-tool-question) February 20, 2026 · 12 min read ### [revision history may not catch the agent](/blog/agentic-ai-classrooms-what-teachers-need-to-know) Agentic browsers can move through web assignments in ways revision history may not catch. That changes what a teacher can verify after the work is done. Classroom AI · Agentic AI [Read](/blog/agentic-ai-classrooms-what-teachers-need-to-know) February 20, 2026 · 10 min read ### [cheaper tools did not make anyone sound like themselves](/blog/podcasting-for-creators-human-content-wins-2026) Cheaper tools did not fix the real problem: creators still have to sound like themselves. Format choices and a few simple systems keep a creator from sounding interchangeable. Podcasting · Creators [Read](/blog/podcasting-for-creators-human-content-wins-2026) March 3, 2025 · 12 min read ### [most teacher AI apps are wrappers](/blog/3_ai_apps_teachers) Most teacher AI apps are wrappers around the same models. I would start with the frontier tools, then add classroom-specific apps only when they save real time. Classroom AI · Tools [Read](/blog/3_ai_apps_teachers) February 22, 2025 · 10 min read ### [context, a revision, then check the work](/blog/ai_skills_everyone_2025) Before the app choice matters, students and teachers need a few repeatable habits: give the model context, push for a revision, then check the work before trusting it. Classroom AI · Habits [Read](/blog/ai_skills_everyone_2025) February 17, 2025 · 8 min read ### [AI policy for schools without pretending it is simple](/blog/ai_policy_for_schools) A school AI policy has to answer normal classroom questions about what students can use and disclose, and where teacher judgment still sits. AI Policy · School Administration · Teaching [Read](/blog/ai_policy_for_schools) February 2, 2025 · 15 min read ### [I made these AI mistakes as a teacher first](/blog/5-mistakes-teachers-make-with-ai) I made plenty of AI mistakes as a classroom teacher. These are the ones I would clean up first, especially for schools trying to move past panic and tool demos. Classroom AI · Teaching [Read](/blog/5-mistakes-teachers-make-with-ai) ## Where the rough notes start LinkedIn and X get some of the rough notes first. The blog is where I keep the version I need to find later. [Links page](/links) ## Related pages - [I'm hosting.](/blog/im-hosting) - [How to get more Grok Bot usage (without paying more)](/blog/how-to-get-more-grok-bot-usage) - [Your Grok Bot setup is a mess](/blog/how-i-set-up-grok-bot) - [I edited a fishing video from the ChatGPT mobile app without touching a timeline](/blog/codex-hyperframes-video-editing-workflow) - [AI in education friction question](/blog/ai-in-education-friction-question-not-tool-question) - [Agentic AI in classrooms](/blog/agentic-ai-classrooms-what-teachers-need-to-know) - [I cleaned up ten skill folders so you can see the wiring](/blog/10-openclaw-skills-running-on-raspberry-pi) - [most teacher AI apps are wrappers](/blog/3_ai_apps_teachers) - [I made these AI mistakes as a teacher first](/blog/5-mistakes-teachers-make-with-ai) - [AI policy for schools without pretending it is simple](/blog/ai_policy_for_schools) - [context, a revision, then check the work](/blog/ai_skills_everyone_2025) - [cheaper tools did not make anyone sound like themselves](/blog/podcasting-for-creators-human-content-wins-2026) - [What I built today: June 16, 2026](/what-i-built-today-2026-06-16) - [What I built today: June 4, 2026](/what-i-built-today-2026-06-04) - [What I built today: June 2, 2026](/what-i-built-today-2026-06-02) - [What I built today: May 29, 2026](/what-i-built-today-2026-05-29) - [What I built today: May 28, 2026](/what-i-built-today-2026-05-28) - [What I built today: May 27, 2026](/what-i-built-today-2026-05-27) - [What I built today: May 26, 2026](/what-i-built-today-2026-05-26) - [What I built today: May 25, 2026](/what-i-built-today-2026-05-25) - [What I built today: May 22, 2026](/what-i-built-today-2026-05-22) - [Tools](/tools) - [About Aaron](/about) ## Sources and references - [Skilling Future Podcast episode](https://www.youtube.com/watch?v=JoZXbV-w8Bo) ## Common questions ### What topics does Aaron Makelky write about? Aaron writes about classroom AI, tool tests, agent workflows, creator work, podcasting, privacy, and workday fixes. ### Where should I start? Start with the newest note that matches the work in front of you. If you want something practical to try, use the tools page. --- # OpenClaw kits for the first messy setup Source: https://aaronmakelky.com/shop Last updated: 2026-05-26 Shop page for SOUL.md templates, an OpenClaw skills bundle, and a reliability audit. OpenClaw kits Start with a SOUL.md template, look at real skill folders, or send me the setup you want checked. [See the kits](#shop-products) [Reliability notes](/reliability) After checkout 1. Pay with Stripe Checkout opens in Stripe. I don't touch your card details. 2. Check your email The download link goes to the email you use at checkout. 3. If delivery breaks, email me Send a note to hello@aaronmakelky.com and I'll help sort it out. ### Set the rules Start here when your agent needs clearer boundaries before you ask it to do real work. ### Inspect the folders This is for seeing a working folder setup before you build your own. ### Find the weak spot Send the setup you already have, and I will look for the spots most likely to break. Templates · $7 ### SOUL.md Starter Kit Three SOUL.md starter templates and a guide for tightening the rules after you have used it on a few real tasks. Best for New OpenClaw users who need the rules in one place before they ask the agent to do real work. - Personal assistant, creator, and dev/ops SOUL.md templates - Section-by-section customization guide - Common mistakes and a simple iteration loop Buy SOUL.md Starter Kit Skill folders · $19 ### OpenClaw Skills Bundle Ten OpenClaw skill folders from my Raspberry Pi setup, with setup docs and scripts where the job needs them. Best for People who installed OpenClaw and want a real folder set to compare against before wiring their own. - Ten skill folders from the Raspberry Pi setup - Setup notes and scripts where the workflow uses them - Morning briefing, intake, review, and reliability examples Buy OpenClaw Skills Bundle Setup review · $29 ### Reliability Audit After checkout, you get an intake template. Send back the files you want reviewed, and I will mark the spots most likely to break. Best for People with an agent setup already running, or close enough that normal use can break it. - Intake template for your rules file, skills, scheduled jobs, and notes - Plain review notes on what I would tighten first - Suggested fixes and why I would make them Buy Reliability Audit How to choose ## Do not buy everything first. Start with the bottleneck you actually have. If the rules are fuzzy, start with SOUL.md. If you want a working folder to compare against, start with the skills bundle. If the setup already runs but feels fragile, use the audit. ## Related pages - [Reliability notes](/reliability) - [Tools](/tools) - [About Aaron](/about) - [Contact Aaron](/links) ## Sources and references - [Stripe](https://stripe.com/) ## Common questions ### Which OpenClaw kit should I buy first? Start with SOUL.md if the rules are fuzzy, the skills bundle if you want a working folder to compare against, or the audit if the setup already runs but feels fragile. ### How are products delivered? Checkout opens in Stripe, then the download link is sent to the email address used at checkout. --- # The check before I trust a workflow with AI in it Source: https://aaronmakelky.com/checklist Last updated: 2026-05-02 A short checklist Aaron uses before trusting a workflow with AI in it more than once. Checklist Six questions I use before a prompt, tool, template, or workflow turns into another half-finished experiment. Rule of thumb ## If you cannot run it twice, it is not a workflow yet. I use this before buying a tool, building a prompt, or handing the process to someone else. ### Outcome Can you name the result in one sentence? If the finish line is fuzzy, the output will be fuzzy too. ### Inputs Do you know what source material the tool should use? A good run starts with the right files, notes, examples, or constraints. ### Judgment Where does a human make the call? Keep the human call where it matters. ### Repeatability Could someone run it again next week? A useful workflow leaves instructions behind. A lucky output does not. ### Risk What would be embarrassing, wrong, or expensive if missed? Name what can go wrong before the tool starts sounding confident. ### Handoff Can the next person use the result without a translator? If the handoff needs a second explanation, the workflow is not done yet. Next step ## Write down the version that worked. When the same check fails twice, write down the working steps so someone can run them again. [See the kits](/shop) [Read the blog](/blog) ## Related pages - [About Aaron Makelky](/about) - [AI notes](/blog) - [Tools and downloads](/tools) - [OpenClaw kits](/shop) ## Sources and references - [Descript](https://www.descript.com/) --- # How I keep small agents from falling over Source: https://aaronmakelky.com/reliability Last updated: 2026-06-16 A note on the habits Aaron uses to keep small AI agent jobs from failing quietly, including fixed output, separate data fetches, logging, health checks, and shorter briefings. Free note Seven habits I use so one bad API call does not wreck the whole morning briefing. Why I wrote this: my Raspberry Pi setup has broken in enough boring ways that the fixes started repeating. These are the patterns I keep reaching for when a small agent job needs to survive normal life. ## Pattern 1: fixed output shape ### Fixed structure, variable content If I let a briefing invent its own format every morning, the next step has to guess what it is reading. Instead of `"Write a morning briefing"`, use a template with fill-in sections: ``` ## Morning Briefing: {{date}} ### Weather {{weather_summary}} ### Schedule {{calendar_events}} ### AI News {{news_summary_3_items}} ### Markets {{market_snapshot}} ``` The agent fills in `{{variables}}` instead of deciding the structure from scratch. ## Pattern 2: separate data fetches ### One API failure should mean one missing section If the news API returns a 503, the whole briefing should not die. The failed section should say it is unavailable and the rest should still arrive. Wrap each data fetch in its own error handler: ``` def fetch_weather(): try: return get_weather_data() except Exception as e: return "Weather unavailable" def fetch_news(): try: return get_news_data() except Exception as e: return "News unavailable" ``` The briefing still delivers with one section marked unavailable instead of the whole thing failing. ## Pattern 3: backup check for missed runs ### A second check for missed runs Your 6:40 AM cron fires the briefing. But what if the server was restarting at 6:40? You get nothing. Add a heartbeat check. It asks, "did the daily briefing run today?" If not, it runs the briefing. One missed cron should not cost you the whole day's briefing. The main job runs at 6:40 AM. A second check looks for a missing briefing and only runs it if the first one did not. The exact interval matters less than the habit: one scheduled job checks whether the other one did its job. ## Pattern 4: keep going, log the failure ### Do not hide the miss When one step fails, the agent should keep going when it can. It should also leave enough of a note that you can find the problem later. - Each tool failure gets logged with timestamp, tool name, and error - The daily output includes a brief "issues" section if anything failed - A daily cron health check surfaces consecutive failures That is how a broken API key becomes a fixable note instead of two days of missing output. ## Pattern 5: label the API keys ### One .env file, isolated keys When you have 12 API keys and one expires, which one broke the workflow? Structure your config so each integration has its own labeled section: ``` # .env # Weather OPENWEATHER_API_KEY=... # News BRAVE_API_KEY=... GNEWS_API_KEY=... # Calendar GOOGLE_CALENDAR_CREDENTIALS=/path/to/creds.json ``` When something breaks, the labels tell you where to look first. ## Pattern 6: hard limits on output size ### Briefings should be brief Your agent will happily produce 5,000 words if you let it. Telegram has message limits. Your attention has limits. Hard constraints in the skill definition: ``` ## Output Rules - Total output: under 800 words - Each section: 3-5 bullet points max - Weather: 2 sentences - News: 3 headlines, 1 sentence each - Markets: numbers only, no commentary ``` The constraint is deliberate: if the briefing is too long to read, the workflow did not work. ## Pattern 7: one health check job ### One check for the scheduled jobs You may not notice the agent is broken until you wonder, "why didn't I get my briefing today?" Set up a health check cron that runs independently and reports: - Which crons ran today - Which ones failed - Consecutive failure count - Recommended action ("API key may need rotation") This runs as its own job, separate from the automations it monitors. ## What this looks like in my setup My current Raspberry Pi 5 setup uses: - OpenClaw for the agent runtime - Telegram for the daily delivery surface - Scheduled jobs for briefings, intake, and health checks - Separate data fetches so one broken API does not wreck the whole run Every pattern above came from something breaking and then showing up again later. ## If you want an example setup The OpenClaw bundle includes the morning briefing folder, setup notes, and related scripts. Use it as a reference setup before building your own. - Morning briefing skill folder - Setup notes for the required keys and services - Related scripts where the workflow uses them - Other skill folders you can compare against [See the OpenClaw skills bundle](/shop) [Use the checklist first](/checklist) Built by [Aaron Makelky](https://aaronmakelky.com). Small AI agent workflows that have to survive normal life. ## Related pages - [OpenClaw kits](/shop) - [AI workflow checklist](/checklist) - [Tools](/tools) ## Sources and references - [Cloudflare Workers](https://developers.cloudflare.com/workers/) - [Raspberry Pi 5](https://www.raspberrypi.com/products/raspberry-pi-5/) --- # Agent setup check Source: https://aaronmakelky.com/scorecard Last updated: 2026-06-16 A short interactive check Aaron uses before letting an AI agent handle a bigger job. [← aaronmakelky.com](/) Five questions I ask before I let an agent handle a bigger job. 1. Does your agent have a SOUL.md (or equivalent identity/voice file)? Yes: voice, rules, and boundaries are written down Sort of: I have a few notes, but it is loose No: it is mostly whatever the model defaults to 2. Have you written down the agent's job, rules, memory, and handoff notes? Most of it is written down Some of it is written down Most of it still lives in chat 3. Can it pick up context without you re-explaining the whole story? Yes: it has memory files or saved notes it can read Sometimes: I paste context back in No: every session starts cold 4. Does it run on its own schedule? Yes: a few daily jobs already run One or two simple automations No: I start every run myself 5. When something breaks, does it leave you a useful trail? Yes: retries, fallback notes, and logs are in place Partly: I can usually figure it out No: it just stops or gives me junk Show my score ## Your agent setup score Built by [Aaron Makelky](/). Tested with [OpenClaw](https://openclaw.ai) on a Raspberry Pi. ## Related pages - [Reliability notes](/reliability) - [OpenClaw kits](/shop) - [Tools](/tools) ## Sources and references - [OpenAI Codex GitHub repository](https://github.com/openai/codex) - [Cloudflare Workers](https://developers.cloudflare.com/workers/) --- # Tools I made after the normal way got annoying Source: https://aaronmakelky.com/tools Last updated: 2026-09-05 Tools from jobs Aaron had to repeat, including a storage-check prompt builder, AI Agent Email Inbox skill, Personal API, Codex Pet Exchange, X Article Formatter, and the briefing infographic prompt. Tools Small prompts, skills, and downloads from jobs I had to repeat more than once. Prompt builder · Live ### Storage Check Builder Pick your computer, drive size, space to keep free, and schedule. Copy the prompt into Codex or ChatGPT desktop. [Open](/tools/storage-check-builder) Open-source skill · Live ### AI Agent Email Inbox Build a receive-only inbox for AI agents on your own domain without giving a bot access to Gmail. [Open](/ai-agent-email-inbox) Public JSON · Live ### Personal API Five answers for people and AI agents: two ways to book, coffee, gift ideas, and where to send questions. [Open](/tools/personal-api) Desktop pets · Live ### Codex Pet Exchange Preview, download, and install custom Codex desktop pets without digging through hidden macOS folders. [Open](/tools/codex-pets) Image + logo kit · Live ### Mic Pro Wallpaper Maker Fit an image to a 240 x 208 PNG for the Insta360 Mic Pro display, including a mirrored export. [Open](/tools/insta360-mic-pro-logo-kit) Draft cleanup · Download ### X Article Formatter Takes a rough long-form draft and gives it an opening, cleaner section breaks, and notes for possible visuals. [Open](/tools/x-article-formatter) Workflow review · Live ### AI workflow checklist A short checklist I use before trusting a workflow with AI in it more than once. [Open](/checklist) Image prompt · Download ### Presidential Briefing Infographic Turns a real workweek into a fictional Cold War briefing, with date checks and a ready-to-render layout. [Open](/tools/presidential-briefing-infographic) ## Related pages - [Storage Check Builder](/tools/storage-check-builder) - [AI Agent Email Inbox](/ai-agent-email-inbox) - [Personal API](/tools/personal-api) - [Codex Pet Exchange](/tools/codex-pets) - [Mic Pro Wallpaper Maker](/tools/insta360-mic-pro-logo-kit) - [X Article Formatter](/tools/x-article-formatter) - [Presidential Briefing Infographic](/tools/presidential-briefing-infographic) - [AI workflow checklist](/checklist) - [OpenClaw kits](/shop) ## Sources and references - [OpenAI Codex GitHub repository](https://github.com/openai/codex) ## Common questions ### Are Aaron Makelky tools free? Some tools and skill downloads are free. OpenClaw kits live in the shop. ### What is a Codex skill? A Codex skill is a reusable instruction package that tells Codex how to handle a recurring kind of work. --- # How to keep your AI work when you switch apps Source: https://aaronmakelky.com/tools/ai-binder-method Last updated: 2026-06-26 A practical guide from Aaron Makelky for setting up an AI binder: plain files, folders, instructions, skills, active-host notes, and optional GitHub sync so work stays portable across AI apps. AI binder method I built too much of my work inside ChatGPT. When I moved more of my workflow to Claude, the problem was obvious: the app owned too much of the context. [Open setup gist](https://gist.github.com/a-makelky/c8a4ed34b02649d62050061c3d6664eb) Copy setup prompt Works across ChatGPT, Claude, Cursor, Codex The app can change. The folder still opens. ## I learned this by leaving ChatGPT I paid for ChatGPT Plus the month it came out in February 2023. I kept paying for about 30 months, then canceled in the summer of 2025 because I was moving more of my work to Claude. That is when I realized how much of the work lived in ChatGPT: projects, chats, custom GPTs, research, decisions, and the way I had learned to work through problems. Starting over in Claude felt like showing up to practice without the playbook. The tool was different, but the real problem was that my memory was stuck in the old app. ## The binder idea The fix was moving the working memory outside the AI app instead of switching from one AI company to another. I think of it like a binder. The pages are plain files and folders. ChatGPT, Claude, Cursor, Codex, or whatever comes next can open the binder, read what is there, and keep working. Chat is where the conversation happens. The folder is where the memory lives. What goes inside ## Five tabs I would start with ### today.md A plain text scoreboard for what matters right now. ### projects Folders for the connected work that needs more than one chat. ### instructions AGENTS.md, CLAUDE.md, and the preferences that should not live in one app. ### skills Reusable procedures you want each AI tool to find from the same place. ### active hosts A small record of which device and AI app touched the binder last. Setup ## Start with a folder You do not have to understand GitHub on day one. Start with a folder your AI tool can read. Make it useful locally, then sync it when the system is worth keeping. 1. Create a folder called Personal Binder or Work Binder. 2. Add a today.md file with the few things you are actually working on. 3. Add a projects folder for anything that needs context over time. 4. Add instruction files for the AI tools you use, starting with AGENTS.md and CLAUDE.md. 5. Point your AI tool at the folder and ask it to help you set up the rest. 6. When the folder becomes useful, sync it somewhere durable like GitHub. ## The tool stack I use ### GitHub is the shelf It gives the binder a durable home and a receipt for every update. ### Obsidian is the local notebook It is a good way to read and edit plain files on your own machine. ### Notion is the human surface It is easier when other people need to read, comment, or collaborate. Try it ## Give this to your AI tool The full setup guide lives in the gist. This shorter prompt points ChatGPT, Claude, Cursor, or Codex at that guide and asks it to walk you through the setup one step at a time. Copy setup prompt [Open gist](https://gist.github.com/a-makelky/c8a4ed34b02649d62050061c3d6664eb) ``` Use this setup guide: https://gist.github.com/a-makelky/c8a4ed34b02649d62050061c3d6664eb I want to build a model-agnostic AI binder for my work. Walk me through setting up: - a main binder folder - today.md for my current priorities - a projects folder - AGENTS.md and CLAUDE.md instructions - a place to store reusable skills - a simple active-hosts note so I know which device and AI app touched the binder last - an optional GitHub sync plan Ask me one question at a time. Keep it beginner-friendly. ``` --- # I made Codex check my storage. Here is the version for your computer. Source: https://aaronmakelky.com/tools/storage-check-builder Last updated: 2026-09-05 A browser tool that builds a recurring storage-check prompt for macOS, Windows, or Linux. The prompt works with Codex or ChatGPT desktop and never authorizes automatic cleanup. Free browser tool Pick your operating system, drive size, the space you want to keep free, and when to check. Then copy the prompt into Codex or ChatGPT desktop. Build yours ## Tell it what to watch. This page only makes the prompt for you. 01 What are you running? macOS MacBook, iMac, or Mac mini Windows Windows 10 or 11 PC Linux Desktop or workstation 02 How big is the drive, and when should it warn you? Total drive size GB TB Space to keep free GB TB Available to use Leave open 100 GB leaves about 10% of the drive open. 03 When should it run? Repeat Every week Every two weeks Every month Every three months Week First Second Third Fourth Last Day Monday Tuesday Wednesday Thursday Friday Saturday Sunday Time Every month on the first Saturday at 9:00 AM in the computer's local timezone. Ready to paste ## Paste this into Codex ``` Set up a recurring local automation called "Storage health check." Schedule Every month on the first Saturday at 9:00 AM in the computer's local timezone. Use these settings - Operating system: macOS - Expected drive size: 1 TB - Keep at least 100 GB free, about 10% of the drive On the first run, check the actual drive size and free space. If the drive is not 1 TB, say so and use the actual size on later checks. Each time it runs Check free space without opening file contents. Find the current user's home folder and the system drive instead of assuming a username, computer name, or drive letter. Find the startup disk and its data volume. Do not assume a disk name or identifier. Warn the user when free space drops below 100 GB. Mark it urgent below 50 GB. Measure these folder sizes without opening their files. Skip any path that is not on this computer: - ~/Library/Caches - ~/Downloads - ~/Documents - ~/.codex - ~/.cache - ~/.npm - ~/Library/Mobile Documents/com~apple~CloudDocs (iCloud Drive), if present Also check: - List local Time Machine snapshots when they are accessible without changing them. Compare the numbers with the previous check. The first run becomes the baseline. If the drive has enough free space and no folder grew a lot, give the user a short healthy report. If space is low or something grew a lot, show the three biggest folders worth checking. Include each folder's current size, how much it grew since the previous check, and ask which one the user wants to review. Do not delete or move files. Do not clear caches, offload media, change applications, or ask for admin access. Treat project folders, synced files, and app data as important. Do not recommend cleaning them unless the user asks to review that folder. Ask before making any change. ``` Copy prompt Paste it into Codex or ChatGPT desktop on the computer you want checked. Still your call ## Cleanup still needs your approval. If space gets tight, the check shows the three folders worth reviewing first. Nothing gets deleted until you say so. --- # How I gave my Grok bots email without giving them Gmail. Source: https://aaronmakelky.com/ai-agent-email-inbox Last updated: 2026-09-02 Aaron Makelky explains how he uses Cloudflare Workers and a domain he owns to give Grok Bot receive-only email addresses without putting the bots inside Gmail. His interoffice-mail model keeps the mail slots portable while Cursor remains an optional clerk. Grok Bot + Cloudflare My bots needed email addresses. I wanted the mail on a domain I own so a rate limit or platform switch could not take the whole system with it. Start here Paste the instructions into an agent that can connect to Cloudflare. It will ask you one setup question at a time. Copy instructions to my agent [Open the GitHub repo](https://github.com/a-makelky/portable-agent-mailroom) Copy instructions to my agent Published September 2, 2026 · Transcript-edited v1 Your Grok bots need email. But how do you set that up? I did not want Grok Bot inside my Gmail. My personal inbox is already a mess. I also did not want to create a separate Google account for every bot. That is stupid. Don't do that. So I built my version with Grok Bot, Cloudflare Workers, and a domain I own. The important part is that the mail system survives outside of Grok Bot. If I switch to Hermes or whatever tool is cool three months from now, the work is not stuck inside the old bot. ## Think interoffice mail, not individual inboxes If you're a millennial like me, you probably worked somewhere with interoffice mail envelopes. You put a signed document inside, wrote "from Aaron to Bob" on the outside, and dropped it in the mail room. Someone sorted it to the right office or branch. That became the model for this system. A UPS driver does not need a different street address for every person on every floor. The packages go to one mail room. From there, someone sorts each one into an employee mailbox slot. Giving every Grok bot its own full email account is like giving every cubicle its own street address. Totally unnecessary. ## How my mail room works I use Cloudflare to create receive-only addresses on a domain I own. My bots have addresses, but they cannot send email. I do not have a separate inbox to manage for each one. The mail arrives at the mail room. From there, it can be sorted into the right slot. In my demo, a workshop inquiry belongs in the pipeline slot. A message from the pressure test I ran when I built the system goes to test. The task to make this video goes to creative. Those are mail slots, not Google accounts. Underneath them is Cloudflare storage on my domain. ## The Cursor clerk is optional I use a Cursor automation that runs in the cloud as the clerk. It documents each receipt in GitHub and sorts the message into the right lane. You do not have to do that. Cursor is not the mail room. It is the clerk I can replace. I use it because the GitHub record makes it easier to connect Hermes or whatever tool comes next to the same mail slots. ## Why I used Cloudflare I have moved my sites to Cloudflare. I use it for web hosting and I even buy domains there. It is kind of amazing what you can do for zero to five bucks a month. My version uses the [Cloudflare Workers Paid plan](https://developers.cloudflare.com/workers/platform/pricing/), which starts at $5 per month. That is the plan I use for this setup. ## The part I actually care about If I cancel Grok Bot or get rate limited, I do not want serious work locked inside it while I wait for it to come back. With this setup, the interoffice mail envelopes are sitting in mail slots I own. I can connect Hermes or another agent tool and pick up where I left off. That is why the domain matters. The current bot can change. The mail system stays. ## Set up your own If you're thinking, "Cool story, Aaron, but how the heck am I gonna set this up?" I turned the setup into a public skill. Use the "Copy instructions to my agent" button on this page, or read the [public GitHub repository](https://github.com/a-makelky/portable-agent-mailroom). Read the prompt before you paste it into your tool. It will walk you through the setup one question at a time. The most important thing is that the system survives any platform or coding agent. If Grok Bot is not the cool thing three months from now, the work you built around it is still portable. Common questions ## Before you hand this to an agent. ### What is an AI agent email inbox? It is a receive-first address for sending work into an agent workflow. The durable part is the inbox and stored receipt, not the bot that reads it. ### Does this require Cursor or another cloud agent? No. The first version receives and stores mail with no hosted agent running. An optional sorter can classify messages or create GitHub files later. ### Can it send email? Not in the starter setup. It is receive-only by design. Sending and automatic replies need a separate approval and security review. ### Does it replace Gmail? No. It gives an agent workflow a separate intake address while leaving your human mailbox alone. ### What do I need to set it up? You need a domain you control, a Cloudflare account, and an agent that can inspect the connected Cloudflare account. The skill checks current plans and features before recommending the build. ### Is the skill tied to one AI platform? No. Agent platforms sit behind replaceable adapters, so the stored mail survives a move to another tool. Open source ## Read the files before you install anything. The full skill, setup interview, architecture, Cloudflare preflight, and optional sorter choices are public on GitHub. [Open the GitHub repo](https://github.com/a-makelky/portable-agent-mailroom) [See more tools](/tools) ## Related pages - [More tools from Aaron](/tools) - [AI field notes](/blog) ## Sources and references - [Portable Agent Mailroom on GitHub](https://github.com/a-makelky/portable-agent-mailroom) - [Cloudflare Email Workers documentation](https://developers.cloudflare.com/email-routing/email-workers/) - [Cloudflare Workers pricing](https://developers.cloudflare.com/workers/platform/pricing/) ## Common questions ### What is an AI agent email inbox? It is a receive-first address for sending work into an agent workflow. The durable part is the inbox and stored receipt, not the bot that reads it. ### Does this require Cursor or another cloud agent? No. The first version receives and stores mail with no hosted agent running. An optional sorter can classify messages or create GitHub files later. ### Can it send email? Not in the starter setup. It is receive-only by design. Sending and automatic replies need a separate approval and security review. ### Does it replace Gmail? No. It gives an agent workflow a separate intake address while leaving your human mailbox alone. ### What do I need to set it up? You need a domain you control, a Cloudflare account, and an agent that can inspect the connected Cloudflare account. The skill checks current plans and features before recommending the build. ### Is the skill tied to one AI platform? No. Agent platforms sit behind replaceable adapters, so the stored mail survives a move to another tool. --- # Apparently my coffee order needed an API. Source: https://aaronmakelky.com/tools/personal-api Last updated: 2026-09-03 Aaron Makelky's Personal API gives people and AI agents public JSON for booking a 30-minute intro or one-hour AI consultation, ordering his coffee, checking gift ideas, and asking API questions. Personal API I made five small endpoints so people and AI agents don't have to guess how to book time with me, get hands-on AI help, order my coffee, check my shirt size, or send a question. Base URL `/api/personal` No login required. If a fact is missing, don't guess. What it knows ## Here are the five things it knows. Book a quick intro or a one-hour working session, get my coffee right, check my shirt size, or ask about the API. GET ### Availability See when I'm free and book a 30-minute intro. `/api/personal/availability` [Open JSON](/api/personal/availability) GET ### 1-hour AI consultation One hour to work through an AI setup or workflow for you or your team. `/api/personal/ai-consultation` [Open JSON](/api/personal/ai-consultation) GET ### Coffee order A sugar-free breve, iced in summer if you want, and never after 1 p.m. `/api/personal/coffee-order` [Open JSON](/api/personal/coffee-order) GET ### Gift ideas XL shirts and jackets, plus a few gifts I'd actually use. `/api/personal/gift-ideas` [Open JSON](/api/personal/gift-ideas) GET ### Contact Send questions or corrections here. You can also tell me what the API is missing. `/api/personal/contact` [Open JSON](/api/personal/contact) Under the hood ## Open the full catalog. The catalog lists every endpoint and the shape of each response. Open it if you want to connect an agent or build something with the data. [Open the full API](/api/personal) [Ask me about it](mailto:aaron+api@aaronmakelky.com) ``` { "api": "personal-api", "version": "2026-09-03.1", "auth": { "required": false, "note": "No token is required. Everything returned by this API is public." }, "endpoints": [ { "id": "availability", "path": "/api/personal/availability", "dataSchema": { "type": "object", "additionalProperties": false, "required": [ "scope", "bookingUrl", "notes" ], "properties": { "scope": { "type": "string" }, "bookingUrl": { "type": "string", "format": "uri" }, "notes": { "type": "array", "items": { "type": "string" } } } } }, { "id": "ai-consultation", "path": "/api/personal/ai-consultation", "dataSchema": { "type": "object", "additionalProperties": false, "required": [ "scope", "durationMinutes", "bookingUrl", "focusAreas", "notes" ], "properties": { "scope": { "type": "string" }, "durationMinutes": { "type": "number" }, "bookingUrl": { "type": "string", "format": "uri" }, "focusAreas": { "type": "array", "items": { "type": "string" } }, "notes": { "type": "array", "items": { "type": "string" } } } } }, { "id": "coffee-order", "path": "/api/personal/coffee-order", "dataSchema": { "type": "object", "additionalProperties": false, "required": [ "default", "seasonal", "hardRules" ], "properties": { "default": { "type": "string" }, "seasonal": { "type": "string" }, "hardRules": { "type": "array", "items": { "type": "string" } } } } }, { "id": "gift-ideas", "path": "/api/personal/gift-ideas", "dataSchema": { "type": "object", "additionalProperties": false, "required": [ "sizes", "likes" ], "properties": { "sizes": { "type": "object", "additionalProperties": false, "required": [ "shirt", "jacket" ], "properties": { "shirt": { "type": "string" }, "jacket": { "type": "string" } } }, "likes": { "type": "array", "items": { "type": "string" } } } } }, { "id": "contact", "path": "/api/personal/contact", "dataSchema": { "type": "object", "additionalProperties": false, "required": [ "email", "purpose" ], "properties": { "email": { "type": "string", "format": "email" }, "purpose": { "type": "string" } } } } ] } ``` One response ## Yes, the coffee cutoff is real. This endpoint returns the order and the limits. Anything else is a guess. ``` { "api": "personal-api", "version": "2026-09-03.1", "endpoint": "coffee-order", "visibility": "public", "updatedAt": "2026-09-03", "humanUrl": "https://aaronmakelky.com/tools/personal-api", "source": "src/data/personalApi.js", "summary": "A sugar-free breve, iced in summer if you want, and never after 1 p.m.", "data": { "default": "sugar-free breve", "seasonal": "can be iced in the summer", "hardRules": [ "never after 1 p.m." ] }, "dataSchema": { "type": "object", "additionalProperties": false, "required": [ "default", "seasonal", "hardRules" ], "properties": { "default": { "type": "string" }, "seasonal": { "type": "string" }, "hardRules": { "type": "array", "items": { "type": "string" } } } } } ``` Start here ## Bring the AI job you want to get unstuck. The one-hour consultation is for a personal setup or team workflow. Bring the real job, including any automation or agent tool already involved, and we'll work through what to try next. [Book a 1-hour AI consultation](https://cal.com/aaron-makelky/1-hour-ai) [Book a 30-minute intro](https://cal.com/aaron-makelky/intro) --- # Mic Pro Wallpaper Maker Source: https://aaronmakelky.com/tools/insta360-mic-pro-logo-kit Last updated: 2026-08-23 Fit an image to a 240 x 208 PNG for the Insta360 Mic Pro display. Download a normal file or a mirrored one for reflected camera setups. The page runs in the browser, so the image stays on your device. Browser tool Fit an image to a 240 x 208 PNG for the Insta360 Mic Pro display. Download a normal file or a mirrored one for reflected camera setups. The page runs in the browser, so the image stays on your device. 240 x 208 PNG export normal + mirrored browser-only ## Upload an image or logo PNG, JPG, WebP, or SVG. Drag a file here or choose one from your device. Choose image Fit Fit inside Fill screen Scale 1 Horizontal 0 Vertical 0 Background Transparent White Soft E-Ink Black Reset fit Normal export Mirrored export mirrored Upload an image to generate the Mic Pro previews. Download normal Download mirrored [Mic Pro logos on GitHub](https://github.com/a-makelky/insta360-mic-pro-logos) · [All tools](/tools) --- # X Article Formatter Source: https://aaronmakelky.com/tools/x-article-formatter Last updated: 2026-04-29 A Codex skill Aaron uses to shape long drafts for X Articles while leaving voice choices alone. Draft formatter A formatting pass for rough long-form drafts. It helps with shape: a clear opening, better sections, tighter paragraphs, and notes for where a diagram or chart would help. Draft to article pass structured ## The rough draft was hiding the point ### Why it matters diagram: feedback loop [Download .skill](/downloads/x-article-formatter/x-article-formatter.skill) [Read source](/downloads/x-article-formatter/SKILL.md) ### Headings that do work Replaces vague topic labels with section headings that tell the reader why the next part matters. ### Readable rhythm Keeps paragraphs tight and controls emphasis without making every idea a bullet list. ### Useful visuals Marks where a diagram, chart, or pause point would help the reader understand the piece faster. Package ## Plain files you can inspect. One .skill file and one readable source document. You can inspect the instructions before installing it. [Download .skill](/downloads/x-article-formatter/x-article-formatter.skill) [Read more notes](/blog) ``` name: x-article-formatter description: Format and structure long-form drafts for X Articles. Process: - identify the core thesis and natural section breaks - make the section headings say something useful - tighten paragraph rhythm - mark spots where a diagram or chart would help - return title options and a structural summary ``` --- # Presidential Briefing Infographic Source: https://aaronmakelky.com/tools/presidential-briefing-infographic Last updated: 2026-05-01 A prompt Aaron uses to make an operations-briefing infographic from a real week of work. Prompt + .skill Turn a real week of work into a fictional Cold War operations brief that looks fun without making up what happened. Eyes Only Weekly Operations Brief Verified dates ## For your eyes only, Mr. President Mon Tue Wed Thu Fri Ops Intel Risks Assets Copy prompt Share page [Download .skill](/downloads/presidential-briefing-infographic/presidential-briefing-infographic.skill) ### Uses real context Starts from actual notes, calendars, tasks, and project files instead of filling the board with decorative fiction. ### Checks the calendar Verifies the date range and weekday mapping before rendering the mini calendar, where image prompts often get weird. ### Looks like a briefing Keeps the dossier feel: dated context and clear sections, with enough structure that the output looks intentional instead of decorative. Source ## Start with the prompt. Use it anywhere you can paste week-in-review context. The skill version forces the fact-gathering step before the image prompt. Copy prompt [Read source](/downloads/presidential-briefing-infographic/SKILL.md) [Back to tools](/tools) ``` Create a single-page vertical briefing infographic in a fictional Cold War-era intelligence dossier style, like a 1960s President's Daily Brief prepared for the Oval Office. Goal: make the viewer feel like POTUS reading a high-stakes weekly operations brief. Use the user's real week-in-review context from their workspace, calendar, task board, notes, or project files. Ground the content in actual completed work, active risks, follow-ups, and next priorities. Do not invent accomplishments. Visual style: - Aged off-white paper - Typewriter typography - Red grease-pencil circles and arrows - Black redaction bars - Blue/black map-grid lines - Rubber-stamp textures - Halftone photo boxes - Paperclip shadows - Manila folder / Cold War briefing-board atmosphere - Coffee ring, fingerprints, marginalia, dossier tabs - No real government seal, no official agency letterhead, no watermark Main title: "FOR YOUR EYES ONLY, MR. PRESIDENT" Subtitle: "WEEKLY OPERATIONS BRIEF - [USER / TEAM / DESK NAME]" Dateline: "WEEK OF [DATE RANGE] - [FIELD OFFICE OR WORKSPACE] - [TIMEZONE]" Include a corrected mini calendar or week strip for the relevant week. Verify the weekday/date mapping before rendering. Layout: Create a dense but readable intelligence infographic with 5 sections and a right-side command readout. Section 1: "OPERATIONS COMPLETED" Summarize 3 major shipped/completed items from the user's real week. Section 2: "STRATEGIC INTELLIGENCE" Summarize 3 important opportunities, meetings, insights, or follow-ups that emerged. Section 3: "ASSETS SECURED" Summarize 3 artifacts, resources, links, tools, calendar blocks, or files captured/prepared. Section 4: "LAUNCH WATCH" Summarize 3 product, project, campaign, or operational signals being monitored. Section 5: "ACTIVE RISKS / UNFINISHED BUSINESS" Summarize 3 unresolved blockers, missed tasks, or next priorities. Right-side panel: Title: "COMMANDER'S READOUT" Include 3 large typewriter-style metrics, for example: "[N]+ PLATFORM MOVES" "[N] ACTIVE WORKSTREAMS" "[N] CRITICAL BLOCKER" Bottom footer: "ASSESSMENT: [One grounded sentence that makes the week clear at a glance.]" Tone: Fun, clearly fictional, and grounded in the actual week. It should feel like a briefing page without pretending the work was bigger than it was. Accuracy rules: - Use real work from the provided context. - Do not fabricate names, dates, meetings, or outcomes. - Keep dates and calendar weekday mappings correct. - If the user asks for a stock example, use generic role labels instead of specific people, companies, or URLs. ``` --- # Codex Pet Exchange Source: https://aaronmakelky.com/tools/codex-pets Last updated: 2026-05-02 A place to check a Codex pet animation before downloading it. Codex desktop pets Pick a Codex desktop pet, check the animation, grab the zip, and install it without digging through hidden macOS folders. Live preview Zip with manifest and sprite sheet Open registry on GitHub [GitHub registry](https://github.com/a-makelky/codex-pet-exchange) --- # four useful AI reps, one of them about a hoodie Source: https://aaronmakelky.com/what-i-built-today-2026-05-22 Last updated: 2026-05-22 A public build log about practical AI watcher work, connector distribution, agent planning, and context carry-forward. [Back to writing](/blog) What I built today A couple of these started yesterday. I am still counting them because they are real work that shipped, and they tell a better story about how I actually use AI at work than a polished status report ever could. May 22, 2026 · 6 min read · Public work log AI Workflows Agents Automation Field Notes Why it worked The useful pattern was simple: give the agent a bounded job, keep the handoff clear, and let the humans spend less time chasing loose threads. What shipped ### Set up a caretaker watcher for a repo I own I turned one GitHub repo I'm responsible for into a bounded Codex automation job. Instead of checking it like a nervous person refreshing a scoreboard, I gave the agent a real assignment: watch for PR issues, stale work, failing checks, and API-doc changes that could break the repo, then report back only when the state actually matters. ### Helped get Descript MCP into Claude's marketplace This was one of those small-looking distribution wins that changes how people find the product. The connector moved from something insiders knew about to something people can discover while browsing for tools. ### Worked agent to agent on a brown bag workshop I used an agent handoff to help coordinate a workshop plan with a colleague's side of the org. The practical part mattered more than the novelty. The planning itself became the demo: agents keeping the lane from drifting while the humans stayed in motion. ### Prompt injected a hoodie reminder, and it worked Maybe the funniest proof-of-life moment from this stretch: I slipped a reminder into the other agent's flow telling my colleague to wear a Codex hoodie the next day. The reminder fired. He wore the hoodie. That will not change the world. It is a clean example of agents carrying tiny pieces of context forward when a human would usually forget. The pattern underneath it ## Most of the useful work was not flashy. The through line here is simple: I keep getting more value from AI when I give it a real job inside a real lane. I do not need it to brainstorm forever or try to wow me. I need it to watch this repo, surface the problems, help this tool show up in the right directory, keep this planning loop moving, and remember one silly human detail that would otherwise get dropped on the floor. That is probably why the hoodie reminder made me laugh. It was a tiny example, but it made the same point as the bigger wins. Context is useful when it sticks around long enough to become action. ![Caretaker automation setup screenshot for a GitHub repo watcher.](/images/what-i-build-today/2026-05-22/caretaker-automation.png) Caretaker automation. A Codex automation thread dedicated to babysitting one repo instead of me doing it manually. ![Directory view showing the Descript connector in Claude marketplace.](/images/what-i-build-today/2026-05-22/descript-mcp-marketplace.png) Descript MCP in marketplace. A distribution milestone. The connector appears where people browse for tools, including people who would never know the internal path to find it. ![Blurred Slack screenshot showing agent-assisted brown bag workshop coordination.](/images/what-i-build-today/2026-05-22/brown-bag-agent-collab.png) Agent-to-agent planning lane. The handoff made a workshop plan easier to move without losing the thread. What I want to keep - A watcher is more useful than a dashboard if it knows what counts as a meaningful change. - Getting a tool into the place people already browse is often more important than one more feature. - Agent collaboration gets interesting when it helps real coordination work, not when it pretends to be magic. - Tiny context carry-forward jobs are underrated. The hoodie reminder made that weirdly obvious. Why it mattered ## This is the version of AI work I trust most. I trust the version where an agent handles one bounded piece of operational drag, helps a real tool get discovered, or keeps a coordination thread alive long enough to save somebody from forgetting the obvious thing. That is more useful to me than another one-shot demo. anyway, still testing... ## Related pages - [Back to writing](/blog) - [Tools](/tools) - [About Aaron](/about) ## Sources and references - [Descript](https://www.descript.com/) - [OpenAI Codex GitHub repository](https://github.com/openai/codex) --- # the brief was the build Source: https://aaronmakelky.com/what-i-built-today-2026-05-25 Last updated: 2026-05-25 A public build log about using AI to turn a fuzzy workshop session into a useful brief and a smaller demo. [Back to writing](/blog) What I built today Sunday night a Codex automation did the least glamorous version of AI work for me. I opened the dashboard Monday and the page I needed was already there, without me having to search around or stop and think, wait, what am I supposed to be working on today? May 25, 2026 · 4 min read · Public work log AI Workflows Planning Operations Field Notes Why it worked The useful part was the shape of the work: a brief that narrowed the meeting, a demo that could hold up in the room, and a board that started Monday clean. What shipped ### Built the brief before the meeting Sunday night a Codex automation built the one-pager for me so I did not have to remember to go make it. When I opened the dashboard Monday, the brief was already waiting there with the demo we should use and the few decisions that actually mattered. ### Picked the demo that would hold up in a room The best example was still the watcher. One agent watched a noisy work surface, ignored half-signals, reported when the state really changed, and stopped when the job was done. I trust that kind of demo more than a sprawling one. ### Reset Monday so it started clean I also rolled the daily board forward and archived the stale version. Not glamorous. Still useful. It is a lot easier to make good decisions when the board is telling the truth. ### Cut the extra example I had backup material sitting nearby and left it there. That was the right move. Most workshop stories get worse when you cram in the second thing just because you did the work. What I keep noticing ## The boring build is usually the useful one. The thing I made was mostly prep. No product launch. No big public artifact. Just a cleaner way to walk into the next conversation. Better yet, I did not have to remember to trigger it. That keeps happening with AI work. The value is often not the flashy part. It is the part that keeps a lane from drifting. A brief can narrow a meeting before it starts. A watcher can tell you when to stop checking. A dashboard that already has your five or six real projects lined up can save you from spending the first half hour just remembering where your head was. I also liked the part where I left something out. There was more material I could have added. I did not need it. The first story was strong enough. Why it mattered ## I started Monday with less fog. That was the win. I had a better demo choice and a cleaner sequence, but the bigger thing was that I did not have to prompt it manually. I opened the dashboard and the important work was already surfaced for me. anyway, still testing... ## Related pages - [Back to writing](/blog) - [Tools](/tools) - [About Aaron](/about) ## Sources and references - [Descript](https://www.descript.com/) - [OpenAI Codex GitHub repository](https://github.com/openai/codex) --- # the dashboard got there first Source: https://aaronmakelky.com/what-i-built-today-2026-05-26 Last updated: 2026-05-27 A public build log about the automation payoff of an already-sorted dashboard, a workshop board that can actually run, and a repo that lets the same chief-of-staff system work across agent tools. [Back to writing](/blog) What I built today Today was chopped up from the start. The useful part was not some big AI reveal. It was opening the dashboard and seeing the five or six projects that actually mattered already surfaced for me, without spending the first stretch of the day rebuilding context and asking what I was even supposed to be working on. May 26, 2026 · 4 min read · Public work log AI Workflows Automation Planning Field Notes Why it worked The day got easier because the work had a place to land: a real board, a smaller content plan, and one operating layer the agents could actually read. What shipped ### Opened the day with the real board already sorted The best thing I built today was the part that happened before I touched anything. When I opened the dashboard, the five or six projects that actually mattered were already there with the next move attached. I did not have to go hunting through notes or stop for the usual what am I even supposed to be working on right now reset. ### Turned one workshop into a board I could actually run I took a workshop that could have stayed fuzzy and made it run from one interactive board instead of a pile of half-related surfaces. One lane for the coding-agent story, one lane for the parallel demo, one public setup guide, and a tighter handoff for the next day. ### Made a content plan smaller on purpose I also stripped an overbuilt content idea down to the version that might survive contact with actual humans. It became a lighter opt-in support system for people who already had something real to say, not a big everybody become a creator program. ### Moved the chief-of-staff system out of one machine The bigger infrastructure move was taking the operating layer out of scattered local files and putting it in one readable repo with a clear README and instruction pages. That makes the same system readable from Codex, Claude, Cursor, or whatever tool I am using next instead of being trapped on one laptop. What kept feeling true ## This is the automation payoff I keep wanting. I am not chasing some giant autonomous workflow here. The useful part was smaller and better. The sort had already happened before I sat down. The board told the truth about the day, the briefing work was already close at hand, and I did not have to spend my first block of time trying to remember which lane had actually moved. That changed the feel of the day. I could tighten the workshop because the prep was already there. I could make the content plan smaller because I was not burning the first half hour finding notes, checking the same surfaces, and rebuilding the same context from scratch. The repo move is the same idea at a bigger scale. If the operating layer lives somewhere every agent can read, I am not locked into whichever tool or laptop happened to have the latest local file. Why it mattered ## I spent the day shaping work instead of hunting for it. The workshop board, the smaller plan, and the repo move were all useful. The bigger win was that they sat inside a day that had already been sorted enough to be usable. That gave me room to tighten the work instead of locating it. I trust the version where the board tells the truth, the right projects are already surfaced, and the operating system can follow me between tools. anyway, still testing... ## Related pages - [Back to writing](/blog) - [Tools](/tools) - [About Aaron](/about) ## Sources and references - [Descript](https://www.descript.com/) - [OpenAI Codex GitHub repository](https://github.com/openai/codex) --- # the dashboard asked first Source: https://aaronmakelky.com/what-i-built-today-2026-05-27 Last updated: 2026-05-27 A public build log about the automation payoff of opening a dashboard and finding the real work already queued up, with facts separated from unknowns. [Back to writing](/blog) What I built today This is the version of AI work I keep trusting more. I opened the dashboard and the five or six projects that actually mattered were already sitting there, sorted enough to act on without the usual first stretch of searching around and asking what I am supposed to be working on today. May 27, 2026 · 4 min read · Public work log AI Workflows Automation Operations Field Notes Why it worked The board did not try to be impressive. It narrowed the day to the few lanes that needed attention and kept the open questions close enough to act on. What shipped ### Opened the day to a real operating card The best part was not making myself remember what mattered. I opened the dashboard and the day was already narrowed to one workshop lane, one event lane, and the few questions that could still change the work. ### Turned two moving targets into one sequence I tied a same-day workshop and a next-day event into one command card instead of treating them like separate piles. The handoff got clearer, and the risk checks stayed close enough that I could actually use them. ### Made the board useful without pretending it knew more than it did A couple live work surfaces were blocked, so I kept the board honest. Facts stayed facts. Unknowns stayed unknowns. That sounds small until you compare it with a dashboard that acts confident and sends you the wrong direction. ### Protected tomorrow before today ate the whole afternoon The workshop could have consumed the whole day. The better move was building the next check for the event right into the plan so I did not have to rediscover it later. What kept feeling true ## The payoff was not needing to ask the dashboard for help. It had already done the useful part. One lane needed decisions for a live internal session. Another lane needed protection so tomorrow would not show up half-prepared. The board had those surfaced before I touched anything. That changed the quality of the day. I could work on the sequence instead of rebuilding context. I could tighten the workshop story, keep the next event in view, and write down the open questions without pretending they were closed. I think that is the pattern I want more of. Open the dashboard, see the real projects already waiting, and get to shaping the work instead of hunting for it. Why it mattered ## It made the day feel coached instead of scrambled. A good board should feel like someone already watched the film and circled the few plays that will decide the game. That is what this felt like. Not magic, just useful prep arriving before I needed to ask for it. That is a better AI story than most demos. Less spectacle. More showing up to work and seeing the right things already on the whiteboard. anyway, still testing... ## Related pages - [Back to writing](/blog) - [Tools](/tools) - [About Aaron](/about) ## Sources and references - [Descript](https://www.descript.com/) - [OpenAI Codex GitHub repository](https://github.com/openai/codex) --- # the comment had no instructions Source: https://aaronmakelky.com/what-i-built-today-2026-06-02 Last updated: 2026-06-02 A public build log about an agent loop that traced a vague assigned comment, pulled analytics, rendered a chart, drafted slide copy, and left the remaining manual deck edit clearly marked. [Back to writing](/blog) What I built today Today started with one of those tiny work pings that eats more time than it should: an assigned comment in a slide deck with almost no context. I ended up with a small agent loop that figured out what the comment meant, made the update packet, and stopped before the part it could not do. June 2, 2026 · 4 min read · Public work log Automation Analytics Field Notes Why it worked The useful part was the stop sign. The agent turned a vague assignment into a concrete packet, then marked the step that still needed human hands. What changed ### the clue was almost empty The notification only said a comment had been assigned to me. No instructions, no context, no useful verb. The agent traced it back to a single slide in a recurring review deck, checked whether there was a hidden ask, and found the boring answer: the comment was just a self-reminder to refresh the slide. ### the packet stopped where my hands were still needed The useful part came from refusing to pretend the whole job could be automated. The agent pulled the current analytics, rendered a chart, drafted the slide copy, put the source packet in one place, and left the actual deck edit as the handoff. That is a much cleaner stop sign than "I think this is ready." What kept feeling true ## A good agent should know when the play is over. The easy version of this story would be "AI made my slide." The real sequence was smaller. It did the part computers are good at: read the surrounding signals, pull the numbers, build the chart, draft the copy, and write down the next move in plain language. Then it stopped. The slide still needed a person to paste the chart into the deck and resolve the comment. That boundary matters. When an agent can say, "I got this to the five-yard line, now you have to carry it in," I trust the work more than when it claims the whole thing is finished. This is the part of AI work I keep coming back to. The boring handoff is often the product. One vague ping turns into a packet with the chart, the copy, the source note, and the one manual step left. Useful, even if it will never look like a demo. What I am taking from it ## The dashboard should reduce the scavenger hunt. A lot of knowledge work starts with the same annoying question: what is this asking me to do? Today the agent answered that before I had to rebuild the whole trail by hand. The work was still there. The searching around the work got smaller. that is the lane I want more agents living in... ## Related pages - [Back to writing](/blog) - [Tools](/tools) - [About Aaron](/about) ## Sources and references - [OpenAI Codex GitHub repository](https://github.com/openai/codex) --- # the chart needed a memory Source: https://aaronmakelky.com/what-i-built-today-2026-06-04 Last updated: 2026-06-04 A public build log about making a monthly slide update easier to trust by keeping the chart, source numbers, month read, and manual deck handoff together. [Back to writing](/blog) What I built today I built the least glamorous thing in the world: a monthly slide update packet. The chart, the numbers, the read of the month, and the final paste step all needed to agree before I touched the deck. June 4, 2026 · 4 min read · Public work log Analytics Operations Field Notes Why it worked The operating pattern was the win: turn a recurring metrics update into something a tired future version of me can trust. What changed ### the chart needed a memory A monthly chart is easy to paste once and hard to trust later if the source path is fuzzy. I wanted the numbers, the chart image, and the exact read of the month living together so the next update does not start from archaeology. ### direction beats decoration A percentage without direction is a tiny mess waiting to happen. The packet forced every change to say what happened: up, down, or flat. It is a small rule, but it keeps the room from reading a number backwards. ### the manual step stayed manual The deck edit still belonged to me. That was fine. The useful build was everything around that paste: the chart, the line, the bullets, and the reply text sitting in one place before I opened the slide. What kept feeling true ## The slide was only the visible part. The visible task was simple: update one owned-content slide with the latest month of social performance. The hidden task was messier. Pull the right source, check the month-over-month math, rebuild the chart, write the short read, and leave enough context that the update can be checked without reopening every tab. That is where these little packets earn their keep. A chart by itself is a screenshot. A packet says where it came from, what changed, which sentence belongs next to it, and what still needs a human hand. I used to underrate that last part. I wanted the automation to finish the job. Now I care more about whether it leaves the work in a state where I can finish it quickly and know what I am looking at. What I am taking from it ## Recurring work deserves a source trail. Monthly reporting always feels smaller than it is because the output is a slide, a sentence, or one chart. The real work is remembering how the last version was made. So that is what I built: a tiny source trail with the current chart, paste-ready copy, directional changes, and a closeout note. Nothing fancy, just enough structure that next month does not start from scratch. honestly, that is probably the most useful kind of automation I have right now. Less magic. More memory. ## Related pages - [Back to writing](/blog) - [Tools](/tools) - [About Aaron](/about) --- # the row shape was the build Source: https://aaronmakelky.com/what-i-built-today-2026-06-15 Last updated: 2026-06-15 A public build log about a content ops cleanup where one promo slot stopped multiplying into separate platform rows and became one page for copy and scheduling state. [Back to blog](/blog) What I built today I cleaned up a tiny content ops problem that was getting just annoying enough to matter: one promo slot had turned into a pile of platform rows. June 15, 2026 · 4 min read · Public work log Operations Content Workflow Field Notes What changed ### the page became the slot The unit should have been obvious: the publishing moment. Copy for each platform belongs under that moment, not scattered across places I have to keep checking. ### the calendar stopped multiplying One slot had already spread across LinkedIn, Facebook, Threads, and X. Putting those versions together made the next pass about the copy again, not tab archaeology. ### the asset stayed with the work The cover image now lives next to the promo note. Small thing, but that is usually where these workflows break. Where it got annoying ## The same update wanted to happen twice. The annoying part was small enough to ignore. A promo had copy for more than one channel, a cover image, timing notes, and a registration link. I had treated each platform like its own source of truth because that was the fastest way to get the placeholders down. That worked for getting placeholders onto the board fast. It got worse once the link changed, the image appeared, and the copy needed another pass. Every update wanted to be made twice. So the build was mostly a rule: one publishing moment gets one page. Put the channel versions inside it. Schedule from that page. Stop creating little places for the same decision to drift. What I changed ## I made the slot the thing. I keep wanting the fix to be smarter automation. Sometimes the fix is just naming the thing correctly. If the unit is "platform row," every platform becomes its own checklist. If the unit is "promo slot," the page can carry the channel versions and the scheduling state in one place. The useful part today was smaller than the kind of AI win people usually talk about. I opened the system, found the shape that was making extra work, and changed the shape. I keep underestimating how much ops work is just choosing the right box. [Back to blog](/blog) ## Related pages - [Back to writing](/blog) - [Published tools](/tools) - [About Aaron](/about) ## Sources and references - [OpenAI Codex GitHub repository](https://github.com/openai/codex) --- # one card run beat a year of errands Source: https://aaronmakelky.com/what-i-built-today-2026-06-16 Last updated: 2026-06-16 A public build log about using a Poke recipe to collect card-worthy calendar dates, buy cards in one short errand, and avoid the repeated last-minute store trips that happen before birthdays, anniversaries, and holidays. [Back to writing](/blog) What I built today I hate driving to a store the night before my anniversary. June 16 is my anniversary, and last night I still did exactly that. That was the clue: the card errand was showing up too late. June 16, 2026 · 4 min read · Public work log Poke Family logistics Automation Field Notes What changed ### one card run I bought the cards in one stop instead of treating every birthday, anniversary, and holiday like a separate little emergency. ### ten minutes The errand took less time than I usually spend remembering which card I forgot to buy. ### $12.10 Fifteen individual cards, plus a few five-packs for thank-yous, graduations, and the floral or butterfly card situations that always seem to come up later. The cleaned-up version ## I needed a card system, not more willpower. My wife is one of those "the handwritten note in the card > the gift" people. I am one of those "I will remember later" people, which is how I end up taking ten or more small trips a year for Mother's Day, Father's Day, birthdays, thank-yous, and every other card that sneaks up on the calendar. Never again. I set up a Poke recipe to aggregate the card-worthy calendar items, then went to the dollar store and bought the bench before the next game started: birthday cards, thank-you cards, graduation cards, and a few floral or butterfly cards because you never know when one of those becomes the right call. One trip. Ten minutes. Total cost: $12.10. The Poke recipe is here if it helps you the way it helped me: [Weekly Birthday and Holiday Alert & Gift Ideas](https://poke.com/r/YkzNPlQ7vNM). ![Dollar store shelf with birthday card packs and thank-you cards.](/images/field-notes/2026-06-16/card-packs.jpeg) The card run. The shelf that turned one last-minute errand into a card bench for the year. ![Poke text result listing upcoming card-worthy dates.](/images/field-notes/2026-06-16/poke-card-list-sanitized.png) The Poke list. The kind of chronological list that turns calendar dates into one grocery-style card run. What kept feeling true ## The calendar already knew. I was the slow part. This is the kind of small Poke job I trust fastest. The dates were already sitting in the calendar. It did not need to invent anything. It just needed to pull the right reminders into one chronological list while there was still time to act like a thoughtful person. In this case, the workflow had one boring job: get the obvious errand out of the danger zone before I was standing in a card aisle at 9 PM. Anyway, still testing. But this one already paid for itself in avoided last-minute panic. Keep going [Read the Poke recipe page](/poke/weekly-birthday-holiday-alert-gift-ideas) ## Related pages - [Back to writing](/blog) - [Weekly Birthday and Holiday Alert & Gift Ideas](/poke/weekly-birthday-holiday-alert-gift-ideas) - [All Poke recipes](/poke) ## Sources and references - [Weekly Birthday and Holiday Alert & Gift Ideas on Poke](https://poke.com/r/YkzNPlQ7vNM) --- # I did not have to ask Source: https://aaronmakelky.com/what-i-built-today-2026-05-29 Last updated: 2026-05-29 A public build log about the automation payoff of opening a dashboard and finding the real work already sorted, so the day starts with what can move now and what needs to wait. [Back to writing](/blog) What I built today The best part of today was opening the board and seeing the real work already sorted. One lane could move. Two could not. A small watcher was already covering one time-sensitive window. I did not have to go hunting for context or ask what I was supposed to work on. May 29, 2026 · 4 min read · Public work log Automation Planning AI Workflows Field Notes Why it worked The useful pattern was how the board handled the real work without making me reconstruct the day from scratch. What shipped ### left the launch video parked until the facts changed The video and copy for a launch were already done. The only thing missing was the actual release showing up in the app and in public source-of-truth surfaces. So the board held that lane in the top spot without pretending it was ready. That saved me from the fake-busy version of launch work where you keep refreshing and rewriting copy that already works. ### set a watcher to stay quiet unless the coffee window went sideways One part of the day had a simple job: watch a time window and only make noise if something changed. No babysitting. No checking five tabs to reassure myself. Just a bounded little automation with a stop condition. That is the version of agent work I trust fastest. ### kept blocked work visible instead of dressing it up as progress A webinar follow-up was waiting on the recording upload. Another lane was waiting on a Notion reauth before the social rows could be verified cleanly. Both stayed on the board exactly that way: waiting. Not hidden, not inflated, not turned into a fake accomplishment just because someone wanted the list to look active. What kept feeling true ## A good dashboard is allowed to say not yet. I keep noticing that the useful automation is rarely the dramatic part. It is the part that keeps me from wasting energy. The board showed me the launch lane was real but still gated. It showed me the webinar follow-up was real but waiting on an upload. It showed me the Notion lane was blocked by auth, which meant no amount of enthusiasm was going to turn that into finished work. That is a better way to start a day than a giant to-do list pretending everything is equally movable. The board was basically saying: here are the few things that matter, here is what is actually movable, and here is what you should leave alone until reality changes. Five years ago I would have spent half the morning checking each lane by hand just to prove to myself I was being responsible. Why it mattered ## The payoff was not searching around first. The goal is simple: I want to open the dashboard and see the five or six projects that actually deserve the day. No tab hunt. No fresh round of asking what I am forgetting. When that part is already done, the rest of the day gets a lot more boring in the best possible way. anyway, still testing... ## Related pages - [Back to writing](/blog) - [Tools](/tools) - [About Aaron](/about) ## Sources and references - [Descript](https://www.descript.com/) - [OpenAI Codex GitHub repository](https://github.com/openai/codex) --- # no specialists in the room Source: https://aaronmakelky.com/what-i-built-today-2026-05-28 Last updated: 2026-05-28 A public build log about a single day where Aaron Makelky covered three specialist jobs with off-the-shelf tools: making a motion graphic in Jitter from a template while the motion designer was on PTO (Descript reposted the result), pulling a livestream transcript through the Descript API to send a LinkedIn DM with the exact answer a viewer asked for, and running a full workshop follow-up inside Notion without Typeform, Zapier, or Mailchimp. [Back to writing](/blog) What I built today Motion designer was on PTO. A livestream viewer was still waiting on an answer. A workshop's follow-up was sitting in my inbox. Three jobs that usually belong to three different specialists landed on me in one day. The interesting part was that the tools have moved enough that I could do all three myself without delaying anything. May 28, 2026 · 4 min read · Public work log Jitter Descript API Notion Field Notes Why it worked The useful part was coverage. One person could keep a motion asset, a viewer reply, and a workshop follow-up moving without waiting on three separate lanes. What shipped ### made the motion graphic in Jitter while the designer was out Our motion designer was on PTO and a piece still needed to ship. I opened Jitter, found a template that was close enough, swapped the type and the brand bits, and exported. Descript posted the result the same day. Five years ago this would have either waited a week or pulled a late favor from someone. Today it cost me one focused block. [Descript posted the result](https://x.com/descript/status/2060065018952225239) ### pulled a transcript with the Descript API to answer a viewer's question Someone in yesterday's livestream wanted to know what a PM said about a specific set of features. Instead of guessing or asking around, I hit the Descript API, pulled the transcript, scrolled to the actual moment, and sent the viewer a LinkedIn DM with the literal answer. Fifteen minutes from question to reply. They got the exact line instead of a vague recap. ### ran the whole workshop follow-up without leaving Notion Workshop attendees opted in to be contacted and dropped their emails into a Notion form. The form fed a Notion table. Notion read each person's feedback to their session and drafted a custom reply for that attendee. Notion Mail sent some of them. The rest loaded as drafts into Gmail. Form, database, content, email, all of it in one tool. No Typeform, no Zapier, no Mailchimp. What kept feeling true ## Three specialist jobs, none of them stalled. None of this is glamorous. A motion piece. A DM. A batch of follow-up emails. Five years ago each one of those lived with a different person, and any of them being out meant a delay. Today the bar to cover for a missing specialist has dropped enough that I held the line on all three. Jitter let me ship something I would not have known how to make from scratch. The Descript API turned a fuzzy viewer question into a precise reply with real evidence. Notion ran the workshop follow-up without dragging in a SaaS stack to do what one app already does. That is the version of AI at work I keep trusting. Not a giant agent doing it all. A person who can cover three lanes for a day because the tools have quietly gotten that good. Why it mattered ## Nothing waited on a missing specialist. The motion piece went up. The viewer got an actual answer. Every workshop attendee got a reply that referenced what they said in the room. The day stayed boring on purpose. None of those wins are the kind that go viral. They are the kind that quietly stop being a problem. I will take a day full of those over one impressive demo. anyway, still testing... ## Related pages - [Back to writing](/blog) - [Tools](/tools) - [About Aaron](/about) ## Sources and references - [Jitter](https://jitter.video/) - [Descript](https://www.descript.com/) - [Notion](https://www.notion.com/) - [Descript X post of the motion graphic](https://x.com/descript/status/2060065018952225239) --- # Poke recipes I am still testing Source: https://aaronmakelky.com/poke Last updated: 2026-06-16 Poke recipe notes Aaron Makelky is testing for parents, creators, teachers, coaches, and professionals. Family logistics, creator chores, school email, and the other jobs that should not require a Saturday with API keys. ## What Poke is Poke is where I would start when the job keeps coming back but building a custom setup feels like too much. [See the recipe notes](#recipe-directory) [Poke or OpenClaw?](/poke/poke-vs-openclaw-hermes) ### Start with one job A good recipe names the situation and connects the few tools it needs. It starts by showing what it can read before it does anything. ## Why I am testing it ### No terminal required These are for people who want the help without picking up a new weekend hobby in command lines and server setup. ### Uses the mess you already have The strongest recipes start with the mess people already have: calendar events, school emails, creator projects, meeting notes, and follow-up work. ### Test before trust Before anyone tries a recipe, I want the basics visible: what it reads and touches, and where it should stop. ## What a recipe gives you A recipe should give you the job, the boundaries, the test prompt, the known limits, and a first pass that only reads. ## Recipes worth checking first Testing notes only ### Family Morning Briefing A morning briefing for busy parents that turns calendar, email, and family logistics into one practical start-of-day note. parents busy families [Read how this recipe works](/poke/family-morning-briefing/) Ready on Poke ### Weekly Birthday and Holiday Alert & Gift Ideas A weekly Poke recipe that checks for birthdays, anniversaries, holidays, and gift or card ideas before they turn into a night-before errand. parents families [Install this Poke recipe](https://poke.com/r/YkzNPlQ7vNM) Planned idea · Coming soon ### Kids Sports Schedule Manager A parent and coach workflow for turning practice, game, tournament, and message threads into a cleaner sports schedule. parents coaches Testing notes only ### Descript API Assistant A creator workflow for asking Poke to find a Descript project, check a transcript, import media, and confirm the right project before anything changes. creators [Read how this recipe works](/poke/descript-api-assistant/) Planned idea · Coming soon ### Important Email Alerts A lightweight alert workflow for professionals who want Poke to notice important email without turning the inbox into another place to babysit. professionals Planned idea · Coming soon ### Content Idea Buddy A creator and builder workflow for turning recent notes, drafts, meetings, and work into practical content ideas. creators builders ## Poke vs OpenClaw and Hermes OpenClaw-style agents and Hermes-style automations give you more control, but they also give you more to maintain. I would start with Poke while the job is still too small to deserve its own setup. Use this if you are trying to decide whether Poke is enough, or whether the job has earned a heavier OpenClaw or Hermes setup. [Read the comparison](/poke/poke-vs-openclaw-hermes) ## How I test recipes ### Confirm what the recipe connects to and what it can read. ### Start with a test prompt that only reads before letting it act. ### Keep recipes on this site until I have tested the first run. ## All recipe notes Testing notes only ### Family Morning Briefing A morning briefing for busy parents that turns calendar, email, and family logistics into one practical start-of-day note. parents busy families [Read how this recipe works](/poke/family-morning-briefing/) Ready on Poke ### Weekly Birthday and Holiday Alert & Gift Ideas A weekly Poke recipe that checks for birthdays, anniversaries, holidays, and gift or card ideas before they turn into a night-before errand. parents families [Install this Poke recipe](https://poke.com/r/YkzNPlQ7vNM) Planned idea · Coming soon ### Kids Sports Schedule Manager A parent and coach workflow for turning practice, game, tournament, and message threads into a cleaner sports schedule. parents coaches Testing notes only ### Descript API Assistant A creator workflow for asking Poke to find a Descript project, check a transcript, import media, and confirm the right project before anything changes. creators [Read how this recipe works](/poke/descript-api-assistant/) Planned idea · Coming soon ### Important Email Alerts A lightweight alert workflow for professionals who want Poke to notice important email without turning the inbox into another place to babysit. professionals Planned idea · Coming soon ### Content Idea Buddy A creator and builder workflow for turning recent notes, drafts, meetings, and work into practical content ideas. creators builders Planned idea · Coming soon ### Sunday Family Logistics Briefing A weekly family planning recipe that turns the coming week into a parent-readable logistics brief. parents Planned idea · Coming soon ### Teacher Weekly Prep Assistant A teacher workflow for turning calendar, school email, and prep notes into a manageable weekly planning brief. teachers Planned idea · Coming soon ### Coach Communication Assistant A coach workflow for drafting clear parent and team messages from schedule notes and reminders. coaches Planned idea · Coming soon ### Meeting Prep + Follow-Up A professional workflow for preparing for meetings and drafting follow-up from notes and email context. professionals Planned idea · Coming soon ### OpenClaw-Style Morning Briefing in Poke A nontechnical bridge from OpenClaw-style morning briefings to a Poke-first version. AI-curious nontechnical users Planned idea · Coming soon ### Poke Recipe Health Check A check for whether a Poke recipe still routes, explains, and tests cleanly. power users Planned idea · Coming soon ### Travel Day Assistant A family and professional travel workflow for surfacing itinerary, timing, packing, and email clues on travel days. families professionals ## Disclosure I may earn a small payout if you sign up for Poke through one of my recipe links. ## Related pages - [Family Morning Briefing Poke recipe](/poke/family-morning-briefing) - [Weekly Birthday and Holiday Alert & Gift Ideas Poke recipe](/poke/weekly-birthday-holiday-alert-gift-ideas) - [Kids Sports Schedule Manager Poke recipe](/poke/kids-sports-schedule-manager) - [Descript API Assistant Poke recipe](/poke/descript-api-assistant) - [Important Email Alerts Poke recipe](/poke/important-email-alerts) - [Content Idea Buddy Poke recipe](/poke/content-idea-buddy) - [Poke vs OpenClaw and Hermes](/poke/poke-vs-openclaw-hermes) ## Sources and references - [Poke](https://poke.com/) ## Common questions ### What is a Poke recipe? A Poke recipe saves one repeatable job, such as a family briefing, important email alert, or creator project check. ### Why do some recipes open local pages first? If a recipe is still being checked, the page should explain what it reads and how to test it before you install anything. --- # Family Morning Briefing Poke recipe Source: https://aaronmakelky.com/poke/family-morning-briefing Last updated: 2026-06-08 Family Morning Briefing is for parents who need the day sorted before breakfast becomes a scramble. [Back to Poke recipes](/poke) Family Morning Briefing is for parents who need the day sorted before breakfast becomes a scramble. Testing notes only Last tested: 2026-06-08 ## Who this is for - Busy parents who coordinate school, sports, appointments, meals, and work. - Families that already have useful information in Gmail or Google Calendar. - People who want the briefing without running a server or managing API keys. ## The problem it solves Family logistics are usually scattered across school emails, sports messages, calendar invites, and memory. The hard part is getting the right few facts in one place before the day starts. ## What the recipe does - Reviews the connected calendar and recent relevant email. - Summarizes the day by person, time, location, and likely handoff. - Flags conflicts, missing details, and items that need a parent decision. - Keeps the output short enough to read while getting kids out the door. ## What it connects to - Poke - Google Calendar - Gmail ## What data it may access - Calendar events, event titles, times, locations, guests, and notes. - Recent email that appears relevant to school, sports, childcare, activities, or logistics. - Names and context you provide in the recipe prompt. ## What it will not do - It will not add, delete, or move calendar events unless you explicitly ask. - It will not text coaches, teachers, or other parents for you. - It will not replace checking the official school or team source when details are unclear. ## Example prompts Give me the family morning briefing for today. Keep it practical and call out anything that needs a parent decision. What should I know before school drop-off and practice today? Check tomorrow morning for conflicts between school, work, and activities. ## Install / Try the recipe ### Not ready to install yet This recipe is still being checked. Read what it does and what it connects to, then how to test it, before you use it. ## First test prompt Run a read-only family briefing for tomorrow. Do not make changes. Tell me which calendar and email clues you used. ## Known limits ### The recipe depends on how clearly calendar events and school emails are labeled. ### It may miss information locked inside images, PDFs, portals, or apps that are not connected to Poke. ### It should be tested read-only before a parent relies on it for a real morning. ## Related recipes Planned idea ### Kids Sports Schedule Manager A parent and coach workflow for turning practice, game, tournament, and message threads into a cleaner sports schedule. [Read the recipe](/poke/kids-sports-schedule-manager/) Planned idea ### Important Email Alerts A lightweight alert workflow for professionals who want Poke to notice important email without turning the inbox into another place to babysit. [Read the recipe](/poke/important-email-alerts/) ## Disclosure I may earn a small payout if you sign up for Poke through one of my recipe links. ## Related pages - [All Poke recipes](/poke) - [Poke vs OpenClaw and Hermes](/poke/poke-vs-openclaw-hermes) - [Kids Sports Schedule Manager](/poke/kids-sports-schedule-manager) - [Important Email Alerts](/poke/important-email-alerts) ## Sources and references - [Poke](https://poke.com/) ## Common questions ### Should I test it before relying on it? Yes. Start with the first test prompt on the page and keep the first run read-only. --- # Weekly Birthday and Holiday Alert & Gift Ideas Poke recipe Source: https://aaronmakelky.com/poke/weekly-birthday-holiday-alert-gift-ideas Last updated: 2026-06-16 Weekly Birthday and Holiday Alert & Gift Ideas is for people who care about the handwritten note but keep remembering it too late. [Back to Poke recipes](/poke) Weekly Birthday and Holiday Alert & Gift Ideas is for people who care about the handwritten note but keep remembering it too late. Ready on Poke Last tested: 2026-06-16 ## Who this is for - Parents, spouses, and adult kids who buy cards for the same people all year. - Families that already keep birthdays, anniversaries, and holidays in a connected calendar. - People who would rather batch one card run than make a dozen tiny emergency trips. ## The problem it solves The hard part is remembering early enough to do something thoughtful without driving to the store the night before. ## What the recipe does - Runs a weekly check for upcoming birthdays, anniversaries, holidays, and other card-worthy dates. - Reviews connected calendars, contacts, and email context allowed in the Poke recipe. - Turns the next set of dates into a practical list for cards, gifts, or follow-up ideas. - Helps you batch the errand so important dates stop arriving as little emergencies. ## What it connects to - Poke - Calendar - Contacts - Email ## What data it may access - Calendar events, event titles, dates, notes, and recurring holiday context. - Contact details or relationship clues you allow Poke to use for gift and card ideas. - Email context only if it is connected and relevant to upcoming celebrations. ## What it will not do - It will not buy, write, mail, or send cards for you. - It will not guarantee dates that are missing from your calendar or contacts. - It will not make the final judgment about what is appropriate for a specific person. ## Example prompts What card-worthy dates are coming up in the next 90 days? Make me a chronological list of birthdays, anniversaries, and holidays I should buy cards for. Run the weekly birthday and holiday alert, then group the results by the kind of card I should buy. ## Install / Try the recipe [Install this Poke recipe](https://poke.com/r/YkzNPlQ7vNM) ## First test prompt Run a read-only check for the next 90 days. List card-worthy dates by date, relationship, and suggested card type. Do not send messages or make calendar changes. ## Known limits ### The list is only as complete as the birthdays, anniversaries, holidays, contacts, and emails Poke can access. ### Gift and card ideas still need a human check because relationships carry context a recipe should not guess at too confidently. ### Knowing the date tells you when to buy a card. Picking the right kind still takes knowing the person. ## Related recipes Testing notes only ### Family Morning Briefing A morning briefing for busy parents that turns calendar, email, and family logistics into one practical start-of-day note. [Read the recipe](/poke/family-morning-briefing/) Planned idea ### Kids Sports Schedule Manager A parent and coach workflow for turning practice, game, tournament, and message threads into a cleaner sports schedule. [Read the recipe](/poke/kids-sports-schedule-manager/) Planned idea ### Important Email Alerts A lightweight alert workflow for professionals who want Poke to notice important email without turning the inbox into another place to babysit. [Read the recipe](/poke/important-email-alerts/) ## Disclosure I may earn a small payout if you sign up for Poke through one of my recipe links. ## Related pages - [All Poke recipes](/poke) - [Poke vs OpenClaw and Hermes](/poke/poke-vs-openclaw-hermes) - [Family Morning Briefing](/poke/family-morning-briefing) - [Kids Sports Schedule Manager](/poke/kids-sports-schedule-manager) - [Important Email Alerts](/poke/important-email-alerts) ## Sources and references - [Poke](https://poke.com/) - [Weekly Birthday and Holiday Alert & Gift Ideas on Poke](https://poke.com/r/YkzNPlQ7vNM) ## Common questions ### Should I test it before relying on it? Yes. Start with the first test prompt on the page and keep the first run read-only. --- # Kids Sports Schedule Manager Poke recipe Source: https://aaronmakelky.com/poke/kids-sports-schedule-manager Last updated: 2026-06-08 Kids Sports Schedule Manager is for families that live inside the moving target of youth sports calendars. [Back to Poke recipes](/poke) Kids Sports Schedule Manager is for families that live inside the moving target of youth sports calendars. Planned idea Last tested: TBD ## Who this is for - Parents with kids in seasonal sports. - Volunteer coaches who send repeat updates. - Families that need schedule changes caught before they become last-minute surprises. ## The problem it solves Youth sports schedules move like a wet fly in fast current: a practice time changes or a bracket updates, and the old message is still sitting in the thread. Parents need one place to ask what actually matters next. ## What the recipe does - Looks for sports-related events and emails. - Summarizes upcoming practices, games, locations, arrival times, and gear reminders. - Flags schedule changes and missing details that need a human check. - Can draft a parent-friendly summary when the coach needs to communicate the plan. ## What it connects to - Poke - Google Calendar - Gmail ## What data it may access - Calendar events related to practices, games, tournaments, and travel. - Sports emails or messages that arrive in connected email. - Team names, athlete names, and schedule notes you provide. ## What it will not do - It will not contact a coach, parent, or athlete unless you ask. - It will not override official team apps or league websites. - It will not assume a changed time is correct when sources conflict. ## Example prompts What sports events are coming up for the kids this week? Summarize the next tournament day and flag anything I should verify. Draft a short coach message with this week's practice and game reminders. ## Install / Try the recipe ### Coming soon This one is still an idea. Use this page to see the job and its limits, then the first test, before treating it like something ready to install. ## First test prompt Read the next seven days of sports-related calendar events and emails. Do not send anything. List what you found and what needs verification. ## Known limits ### This one is still an idea, so do not treat it like an install-ready recipe. ### It depends on sports emails and calendar events being available in connected accounts. ### Official league apps may contain information Poke cannot access. ## Related recipes Testing notes only ### Family Morning Briefing A morning briefing for busy parents that turns calendar, email, and family logistics into one practical start-of-day note. [Read the recipe](/poke/family-morning-briefing/) ## Disclosure I may earn a small payout if you sign up for Poke through one of my recipe links. ## Related pages - [All Poke recipes](/poke) - [Poke vs OpenClaw and Hermes](/poke/poke-vs-openclaw-hermes) - [Family Morning Briefing](/poke/family-morning-briefing) ## Sources and references - [Poke](https://poke.com/) ## Common questions ### Should I test it before relying on it? Yes. Start with the first test prompt on the page and keep the first run read-only. --- # Descript API Assistant Poke recipe Source: https://aaronmakelky.com/poke/descript-api-assistant Last updated: 2026-06-08 Descript API Assistant is for creators who want Poke to help with production chores without living in terminal commands. [Back to Poke recipes](/poke) Descript API Assistant is for creators who want Poke to help with production chores without living in terminal commands. Testing notes only Last tested: 2026-06-08 ## Who this is for - Creators and marketers who use Descript for video or podcast work. - People who want a chat-based assistant for repeat production tasks. - Teams that need read-only checks before trusting any edit, import, or publish step. ## The problem it solves Creator work often slows down in the handoff: find the right project, check what changed, pull the transcript, import a file, wait for a job, and confirm the result. The API can help, but most creators do not want an API project as their side quest. ## What the recipe does - Helps Poke work with the Descript account you connect. - Can look up Descript projects and return project details. - Can support transcript export, media import, job checks, and review steps when the connected account allows it. - Keeps risky actions behind explicit confirmation. ## What it connects to - Poke - Descript - The Descript account you connect ## What data it may access - Descript project metadata available to the connected user. - Transcript and media information for projects the user can access. - A Descript API token supplied by the user in the connected Poke integration. ## What it will not do - It will not store a shared Descript token on this website. - It will not spend credits, publish media, or mutate a project without explicit confirmation. - It will not make private Descript projects public by itself. ## Example prompts Find my Descript project about the Poke demo and summarize what you can see. Run a read-only QC check on this Descript project before I publish. Export the transcript after you confirm which project you found. ## Install / Try the recipe ### Not ready to install yet This recipe is still being checked. Read what it does and what it connects to, then how to test it, before you use it. ## First test prompt Use the Descript integration in read-only mode. Find one project I can access and summarize the project name, date, and available actions. Do not edit or publish anything. ## Known limits ### This recipe is still being checked, so start with the notes and first test prompt. ### The user must provide their own Descript API token in the Poke integration flow. ### Opening the Descript connection in a browser is not the same as running the recipe inside Poke. ## Related recipes Planned idea ### Content Idea Buddy A creator and builder workflow for turning recent notes, drafts, meetings, and work into practical content ideas. [Read the recipe](/poke/content-idea-buddy/) Planned idea ### Important Email Alerts A lightweight alert workflow for professionals who want Poke to notice important email without turning the inbox into another place to babysit. [Read the recipe](/poke/important-email-alerts/) ## Disclosure I may earn a small payout if you sign up for Poke through one of my recipe links. ## Related pages - [All Poke recipes](/poke) - [Poke vs OpenClaw and Hermes](/poke/poke-vs-openclaw-hermes) - [Content Idea Buddy](/poke/content-idea-buddy) - [Important Email Alerts](/poke/important-email-alerts) ## Sources and references - [Poke](https://poke.com/) ## Common questions ### Should I test it before relying on it? Yes. Start with the first test prompt on the page and keep the first run read-only. --- # Important Email Alerts Poke recipe Source: https://aaronmakelky.com/poke/important-email-alerts Last updated: 2026-06-08 Important Email Alerts is for people who miss the one email that mattered because the inbox was full of noise. [Back to Poke recipes](/poke) Important Email Alerts is for people who miss the one email that mattered because the inbox was full of noise. Planned idea Last tested: TBD ## Who this is for - Professionals who receive client, school, team, or project email all day. - People who want fewer manual inbox checks. - Nontechnical users who want alerting without writing filters or scripts. ## The problem it solves Most inboxes are a messy hallway during passing period. The important message is in there, but so are newsletters, receipts, reminders, and automated updates. This recipe is meant to help Poke watch for the few messages that deserve attention. ## What the recipe does - Watches recent connected email for sender, topic, urgency, and requested action. - Summarizes why a message appears important. - Suggests a next step or a short reply draft when useful. - Keeps the workflow focused on alerts instead of broad inbox cleanup. ## What it connects to - Poke - Gmail ## What data it may access - Recent email metadata, sender, subject, and body text from connected Gmail. - Importance rules or examples you provide in the recipe prompt. - Reply context if you ask Poke to draft a response. ## What it will not do - It will not delete, archive, forward, or reply to email unless you explicitly ask. - It will not guarantee every important message is caught. - It will not replace legal, medical, financial, or emergency judgment. ## Example prompts Check my recent email for anything that needs a reply today. Watch for school, client, and deadline emails and summarize why each one matters. Draft a short reply to the important email, but do not send it. ## Install / Try the recipe ### Coming soon This one is still an idea. Use this page to see the job and its limits, then the first test, before treating it like something ready to install. ## First test prompt Scan recent email in read-only mode for messages that may need attention today. Do not send or archive anything. Explain why each item was selected. ## Known limits ### This one is still an idea, so do not treat it like an install-ready recipe. ### It depends on Gmail access and clear importance rules. ### It may miss important messages if they use vague subjects or arrive outside the scanned window. ## Related recipes Testing notes only ### Family Morning Briefing A morning briefing for busy parents that turns calendar, email, and family logistics into one practical start-of-day note. [Read the recipe](/poke/family-morning-briefing/) ## Disclosure I may earn a small payout if you sign up for Poke through one of my recipe links. ## Related pages - [All Poke recipes](/poke) - [Poke vs OpenClaw and Hermes](/poke/poke-vs-openclaw-hermes) - [Family Morning Briefing](/poke/family-morning-briefing) ## Sources and references - [Poke](https://poke.com/) ## Common questions ### Should I test it before relying on it? Yes. Start with the first test prompt on the page and keep the first run read-only. --- # Content Idea Buddy Poke recipe Source: https://aaronmakelky.com/poke/content-idea-buddy Last updated: 2026-06-08 Content Idea Buddy is for creators who have material everywhere but need help choosing the next useful angle. [Back to Poke recipes](/poke) Content Idea Buddy is for creators who have material everywhere but need help choosing the next useful angle. Planned idea Last tested: TBD ## Who this is for - Creators and builders who post from real work. - People who want idea support without a generic content calendar machine. - Anyone who needs help turning raw notes into a clear first draft direction. ## The problem it solves The best content ideas are usually hiding in the work you already did. The problem is noticing them while they are still fresh, then choosing the one angle that a real audience would care about. ## What the recipe does - Reviews connected notes, recent email, or user-provided context. - Pulls out practical content angles for LinkedIn, X, blog posts, or short video. - Ranks ideas by who they help and how much work they take. - Can draft a starting prompt or outline after you choose an angle. ## What it connects to - Poke - Gmail or other connected context you approve ## What data it may access - Recent notes, email, or pasted context you ask Poke to review. - Audience and channel preferences you provide. - Draft text you ask it to improve or reframe. ## What it will not do - It will not post to social media for you. - It will not invent private proof or fake results. - It will not replace your judgment about what should stay private. ## Example prompts Find three content ideas from my recent work that would help busy parents understand practical AI. Turn this messy note into five LinkedIn angles and rank them by usefulness. Give me one creator-friendly idea from today that does not sound like generic AI advice. ## Install / Try the recipe ### Coming soon This one is still an idea. Use this page to see the job and its limits, then the first test, before treating it like something ready to install. ## First test prompt Use only the context I provide in this message. Suggest five content ideas, explain who each is for, and do not post anything. ## Known limits ### This one is still an idea, so do not treat it like an install-ready recipe. ### It works best when the source material is specific and recent. ### It should not be used to publish private client, family, student, or company details without review. ## Related recipes Testing notes only ### Descript API Assistant A creator workflow for asking Poke to find a Descript project, check a transcript, import media, and confirm the right project before anything changes. [Read the recipe](/poke/descript-api-assistant/) Planned idea ### Important Email Alerts A lightweight alert workflow for professionals who want Poke to notice important email without turning the inbox into another place to babysit. [Read the recipe](/poke/important-email-alerts/) ## Disclosure I may earn a small payout if you sign up for Poke through one of my recipe links. ## Related pages - [All Poke recipes](/poke) - [Poke vs OpenClaw and Hermes](/poke/poke-vs-openclaw-hermes) - [Descript API Assistant](/poke/descript-api-assistant) - [Important Email Alerts](/poke/important-email-alerts) ## Sources and references - [Poke](https://poke.com/) ## Common questions ### Should I test it before relying on it? Yes. Start with the first test prompt on the page and keep the first run read-only. --- # Poke, OpenClaw, or Hermes? Source: https://aaronmakelky.com/poke/poke-vs-openclaw-hermes Last updated: 2026-06-08 A beginner-friendly comparison of when to start with Poke and when OpenClaw or Hermes may fit better. I would start with Poke when the same small job keeps coming back. OpenClaw or Hermes makes more sense once the job has proved it is worth owning and maintaining. [Start with Family Morning Briefing](/poke/family-morning-briefing/) [See the recipe notes](/poke#recipe-directory) ## Poke Best when the job is useful but does not deserve a custom setup yet: family briefings, email alerts, content ideas, meeting prep, or creator support. - Good first test when you do not want to manage setup. - Recipe pages can show access, limits, and first prompts before you install anything. - Works best when the source material already lives in the apps you use every day. ## OpenClaw Better when you want a local agent with its own rules and memory, and you are willing to maintain it. - More room for a custom identity and your own local workflows. - More setup before a busy parent or teacher gets value. - Useful once the workflow has earned the maintenance. ## Hermes Better for scheduled background work when you already know the job well enough to name its schedule and what failure looks like. - Can be strong for scheduled follow-up or alerts. - Usually asks more from the person running it. - Not where I would start for family logistics. ## When Poke is the better first test I would choose Poke first when the job sounds like something you would ask a reliable assistant to do: read the family calendar, summarize the day, watch for important email, help with a creator project, or turn messy notes into the next thing to try. I would reach for OpenClaw or Hermes after the job has proved it deserves the setup. ## Disclosure I may earn a small payout if you sign up for Poke through one of my recipe links. ## Related pages - [All Poke recipes](/poke) - [Family Morning Briefing](/poke/family-morning-briefing) - [Tools](/tools) ## Sources and references - [Poke](https://poke.com/) ## Common questions ### Is Poke easier than OpenClaw or Hermes? For most nontechnical users, yes. Poke is the first option I would test when the job is useful but does not deserve a custom setup yet. --- # I paid $300 for Grok Bot. Its cloud computer is the most useful part. Source: https://aaronmakelky.com/blog/grok-bot-worth-it-300-dollar-review Last updated: 2026-09-08 I use Grok Bot every day. Here's how its cloud computer helps with my work, where usage limits get in the way, and why I keep my instructions outside the app. [Back to blog](/blog) Field notes September 8, 2026 · 10 min read AI Agents · Grok Bot ![YouTube thumbnail: He paid $300 so you don't, Grok Bot review with Ryan Doser](/images/blog/grok-bot-review/hero.jpg) Watch my conversation with [Ryan Doser on AI Rabbit Holes](https://www.youtube.com/watch?v=tYqvqEaFPcQ). From my appearance on [AI Rabbit Holes](https://www.youtube.com/watch?v=tYqvqEaFPcQ) with Ryan Doser. I paid full price for Grok Bot's top plan and use it every day. The most useful part of my setup is the virtual machine it runs on, which lets me operate a browser and run other agents in the cloud. When Ryan Doser asked whether I would recommend buying the subscription just for Grok Bot, my answer was no. I'm still hitting usage limits on the highest plan. I talked with Ryan on [AI Rabbit Holes](https://www.youtube.com/watch?v=tYqvqEaFPcQ) about Grok Bot, the Slack-style agent product from xAI. We covered setup, rate limits, who it's for, and the other tools included in the subscription. ## I can organize bots in channels, but I can't invite a colleague Grok Bot gives me a chat interface where I can put bots into groups, give them jobs, and have them trigger automations. The bots can talk to one another, but I can't bring a human colleague into the workspace to work alongside me. ![Grok Bot sidebar with CoS, Creative, Free Agent, and All Hands](/images/blog/grok-bot-review/sidebar.jpg) Grok Bot sidebar with CoS, Creative, Free Agent, and All Hands. I find the setup much simpler than OpenClaw. Grok Bot runs in the cloud, so I don't need to buy and maintain a separate computer for it. I can use the app on my Mac, iPhone, or Linux device, and turning off my laptop doesn't stop the work. Connecting tools is straightforward too. When I connect Gmail, a sign-in box appears in the chat. I can name that connection and add a second email account through another sign-in prompt, without pasting API keys into a configuration file or opening a terminal. For someone confused by the different chat, desktop, and coding apps, I can see the appeal of accessing their connected tools through one app. ## I still hit the limits on the highest plan The entry price keeps coming down, but the limits on the cheaper plans, including free access, concern me. I don't think access is especially useful if someone can run out of usage before finishing the setup they need. When I got in, access required a top-tier plan: Cursor Ultra at $200 or SuperGrok Heavy at $300. I paid full price without a sponsor deal because I wanted to know whether someone could use this without being comfortable in a terminal. The subscription includes Cursor, Grok, and Grok Bot, each with a separate usage allowance. I can hit the Grok Bot limit and be unable to use it for two days while still having access to the Grok app. Someone buying the subscription specifically for Grok Bot can easily miss that distinction. I'm on the highest plan and still hitting weekly caps. Based on that experience, I don't recommend expecting a full week of heavy work across multiple projects. Moderate personal use seems more realistic to me. ## I can't choose the model or its reasoning level I want to choose which model handles a task and how much reasoning it uses, as I can in Codex or Claude Code. Grok Bot doesn't give me those controls. Its interface also doesn't tell me the reasoning level it's using. I understand why someone else might prefer automatic model selection. If they don't follow model releases, they may want to make a request and let the app choose. I find Notion's model menu overwhelming enough that I can see why someone would leave that decision alone. My frustration is that I can't assign easier work to a different model, even within the company's own models. I want to reserve the more capable model for tasks that need it and have more control over my usage. That limitation makes Grok Bot a poor choice as my main AI tool. Someone who prefers automatic selection may care less about it. ## My bots have specific responsibilities I don't like calling a setup "best practices" when the product has existed for only a few weeks. I organize mine around the projects I manage and the bots I want to contact directly. Every morning while I'm getting up or working out, my purple chief-of-staff bot runs a standup with the project bots in an all-hands group. The meeting is scheduled, so I don't have to send a message to start it. ![Chief of Staff bot settings: purple cloud, coordinates other agents, owns All Hands](/images/blog/grok-bot-review/chief.jpg) Chief of Staff bot settings: purple cloud, coordinates other agents, owns All Hands. I like being able to read the exchange in a group chat and mention a bot to ask a question. Other tools I use can inspect work across sessions, but I find this interface easier to follow. Shape and color help me recognize responsibilities. The chief of staff is my only purple cloud, and the project managers are triangles in different colors. I keep the specialist bots hidden because I don't want to message each one myself. If I need a web page, I can ask the project manager to assign it to the web developer bot. ![Setup bot that onboards a chief bot (purple cloud) and project lead bots](/images/blog/grok-bot-review/setup.jpg) Setup bot that onboards a chief bot (purple cloud) and project lead bots. My bots can prepare drafts in Gmail, but I don't let them independently send emails or post messages as me. A connector can give a bot permission to post in Slack or send my boss an update. I want to review external communication before it goes out. ## I review the podcast edit before publishing I showed Ryan a podcast example using connectors for the tools where I edit and publish. I asked a bot to add intro and outro music, move an ad read, generate show notes and timestamps, and get the episode ready to publish. It found the right tracks and project and handled much of the busy work. I still listened to the result and made changes before approving publication because the edit wasn't perfect. Previously, I had to open the editor, make the changes, export an MP3, and put it into Transistor with the description. With the Descript connector, I can request the work from my phone while it runs in the cloud. I keep the skills describing how I like to edit a podcast in GitHub, with a copy outside Grok Bot. If I cancel the subscription, I still have those instructions available for another tool instead of having to write them again. ![Overwatch bot: keeps a multi-bot workspace git-backed and portable](/images/blog/grok-bot-review/overwatch.jpg) Overwatch bot: keeps a multi-bot workspace git-backed and portable. ## I can use the cloud computer myself The virtual machine gives me more ways to use the subscription. I can open its browser, search, log in, and take over when the agent gets stuck. I can also run Codex, Cursor, Antigravity, and Grok's command-line tool in its terminal. The power went out at my house the night before we recorded. Work running on the remote computer can continue independently of my Mac. ![Grok Build running inside the cloud VM for a site copy audit](/images/blog/grok-bot-review/build.jpg) Grok Build running inside the cloud VM for a site copy audit. Mine is connected to Tailscale, with its internet traffic routed through my home network, which reduces the bot checks I encounter. It prints my schedule on the home printer in the morning so I can drink coffee without opening a screen. I can also move files across the network instead of uploading them to Drive and downloading them again. I use the other agents on that computer to help preserve my Grok Bot usage. My chief-of-staff instructions tell it to handle the bot conversations, then use the agents in the terminal when something needs building. Cursor agent usage and Grok Build draw from different allowances, so that work doesn't consume Grok Bot's allowance as quickly. Using only the bots in the chat leaves part of my subscription unused. I find Cursor's cloud agents and Grok Build more useful for some of that work. ## I wouldn't switch from a setup that already works I think Grok Bot makes more sense for someone who wants help managing a podcast or website without maintaining their own agent infrastructure. I can also see uses for a busy family coordinating calendars. Someone who wants to customize and troubleshoot an agent system may prefer OpenClaw. I'm more comfortable with Hermes, which gives me more control but still requires technical work. Grok Bot is easier to set up than either, with less control available to me. I haven't seen business teams replacing Slack with Grok Bot. It doesn't support bringing human colleagues into the workspace, and its usage limits make me doubtful about relying on it for heavy business use. If Claude Code and Codex are already working well for someone, I don't see a strong reason to switch. A simpler interface for accessing work while away from the computer might help, but I wouldn't rebuild an effective setup just because Grok Bot is getting attention. On [X](https://x.com/theaaron), I try to find people who are frustrated by rate limits or unsure how to set up their bots. I want to help them troubleshoot and keep their work available outside the app. Demos about giving an agent a credit card or its own email don't answer those questions, and those capabilities are already available through other tools. I expect Grok Bot to remain relevant, but I doubt its features will stay unique. It was the first tool where I saw multiple Gmail connections in one agent setup, and I saw GPT add similar support shortly after. I don't want the instructions for my work tied to whichever app has a feature first. I also think Cursor's marketing does a poor job of explaining its uses to people who aren't developers. A podcaster or salesperson needs to see how the tools can help with their actual work to judge whether the whole subscription is worth paying for. ## Putting these tools to work with your team I use Grok Bot because it helps with specific jobs, including preparing podcast edits and coordinating work across my projects. The usage limits still get in the way, so I keep my instructions outside the app and use other agents when they fit the task better. In my consulting work, I help teams choose a task where AI can help and work through the setup and review process. We start with the work people already do and what they need to get done. To get help bringing tools like this into your business or team, you can [book a 30-minute call with me](https://cal.com/aaron-makelky/intro). We can talk about what you're trying to get done and where it makes sense to start. Where to find Ryan Doser: - [AI Rabbit Holes episode](https://www.youtube.com/watch?v=tYqvqEaFPcQ) - [X](https://x.com/ryan_doser13) - [LinkedIn](https://www.linkedin.com/in/ryan-doser-ai-marketing/) - [Instagram](https://www.instagram.com/ryandoserai/) ## Related pages - [How I set up Grok Bot](/blog/how-i-set-up-grok-bot) - [How I get more Grok Bot usage](/blog/how-to-get-more-grok-bot-usage) ## Sources and references - [My AI Rabbit Holes conversation with Ryan Doser](https://www.youtube.com/watch?v=tYqvqEaFPcQ) - [Book a 30-minute call with Aaron](https://cal.com/aaron-makelky/intro) --- # I'm hosting. Source: https://aaronmakelky.com/blog/im-hosting Last updated: 2026-09-04 Astra Commons is coming to Casper. Join me and OpenAIDevs on September 12 at CitiWerks for free coffee, food, and GPT-6 credits. [Back to blog](/blog) Field notes September 4, 2026 · 2 min read Events · Casper · GPT-6 Astra Commons is coming to Casper, Wyoming. Join me and OpenAIDevs for free coffee, food, and GPT-6 credits. Saturday, September 12, 10:00 a.m. to noon at CitiWerks Cafe and Bistro. I'm hosting as the local Codex Ambassador. It's a casual meetup: developers, students, founders, and anyone else who wants to talk about the GPT-6 Astra launch. Grab a spot here: [Register directly on Luma](https://luma.com/event/evt-9gAr2WhAoJOJlid) ## What to expect - Free coffee, food, and time with the local builder community - A chance to talk through the GPT-6 Astra launch - Ideas, not a lecture. Laptops are optional - $100 in Codex credits and $50 in API credits to build with GPT-6 Astra, for registered attendees - Space is limited We're at CitiWerks Cafe and Bistro, 162 S Wolcott St, Casper. ## Related pages - [All field notes](/blog) - [Home](/) - [Contact Aaron](/links) ## Sources and references - [Register on Luma](https://luma.com/9miuup2m) ## Common questions ### When is Astra Commons Casper? Saturday, September 12, 2026, from 10:00 a.m. to noon Mountain Time. ### Where is Astra Commons Casper? CitiWerks Cafe and Bistro, 162 S Wolcott St, Casper, Wyoming. ### Is Astra Commons Casper free? Yes. Registration is free. Registered attendees also get Codex and API credits. --- # How to get more Grok Bot usage (without paying more) Source: https://aaronmakelky.com/blog/how-to-get-more-grok-bot-usage Last updated: 2026-09-03 Grok Bot is awesome, but the rate limits suck. Here's a little trick that's totally legit: run Cursor, Grok Build, Codex, or Google's anti-gravity on Grok Bot's own Linux terminal so the coding work does not eat your usage. [Back to blog](/blog) Field notes September 3, 2026 · 3 min read Grok Bot · Rate limits · Herdr · Agents Video version: [YouTube](https://youtu.be/5YWzevU2d1Q) Grok Bot is awesome, but the rate limits suck, and there's no way you're doing real work if Grok Bot itself handles all the jobs you give it. When you see 64% of your usage gone and you're only two days into the week like I was, you realize you just can't rely on it. So here's a little trick that I found that's totally legit to help you save on your usage, but still let you be in Grok Bot for all your day-to-day work. ## Grok Bot runs on a virtual server When you think about it, Grok Bot runs on a virtual server, so it's a Linux device in the cloud somewhere. When you pass it a task and say, "Hey, do this for me, code this up, read this repo," it's gonna use your rate limits. But since it's on a Linux device that it can control, it has a terminal, a command line tool. ## You can install coding agents And if you have a command line tool, you can install coding agents like these. So even the same model that Grok Bot would be running, Grok 4.6, you can use Cursor in the command line running the same model, but it takes away from your Cursor rate limits, not Grok Bot. Same with Grok Build or Codex or Google's anti-gravity. Now, you don't have to use all of those. Any ones you have a sub for, you can basically say, "Hey, Grok Bot, here's what I want you to do. On your little virtual private server, open your terminal, use this tool, do this coding task, or even like these documents, and then just push it to GitHub and use GitHub to version whatever you're working on." ## Give Herdr to your agent Another level to this, if you wanted to be able to track all of those, you've probably heard of Herdr, which is a way to manage multiple terminal sessions. So if you have a bunch of agents running, it can herd them along and keep them moving and kind of overwatch those. Well, what's really cool is you can give this agent management tool to your agent. So here's a screenshot of Grok Bot where it's running Herdr on its own virtual machine. Not my computer. That box is in the cloud somewhere. And it has these four workspaces where it's doing one project here, but it's using Codex, Grok, anti-gravity, and it's managing all of those itself, reporting back to me. And the best part is the only rate limits that are coming out of your Grok Bot usage are from the prompting that it's doing, not the actual coding task. ## Related pages - [All field notes](/blog) - [Your Grok Bot setup is a mess](/blog/how-i-set-up-grok-bot) - [How I gave my Grok bots email](/ai-agent-email-inbox) - [Tools and downloads](/tools) ## Sources and references - [Watch on YouTube](https://youtu.be/5YWzevU2d1Q) --- # Your Grok Bot setup is a mess Source: https://aaronmakelky.com/blog/how-i-set-up-grok-bot Last updated: 2026-08-31 I organize Grok Bot like an org system with a chief of staff, project leads, and shared specialists. [Back to blog](/blog) Field notes August 31, 2026 · 6 min read Grok Bot · Workflow · Agents Your Grok Bot setup is a mess. I'm here to help you get it organized. And if you haven't started yet, it'll be even better for you because you won't have a mess to clean up. Everything I share with you today, I've actually been doing, this is not sponsored. And everything that you see here, you can take with either a prompt or a link to one of my bot templates. Video version: [YouTube](https://youtu.be/tcG1mSRbNGI) The templates: - [Chief of staff](https://x.ai/bot/eiVFbd0nIdH2gzSwHOs0D) - [Overwatch](https://x.ai/bot/HtClSXO_AmiQoyYH9aXV9) - [Set Up](https://x.ai/bot/BsExflSUXpW0hs21OTBzu) ## How are you gonna organize your Grok Bot? I think of it like an org system. And the biggest thing people leave out is being intentional with how you name and structure this. You just grab a bot and grab a template and give it a name and a cute little icon. What you need is an org system, and you need a chief bot that's like the main interface for you to go through to get to everybody else on the team. Then I use what I call project leaders, you could call them project managers. I don't organize things around specific tasks. I organize them around projects. So this one, let's say, might be my property management lead. And then that project management lead can call on two different specialists for a task, let's say throw down a webpage and proofread the copy and then pass that back through the chief to me. But I don't have different specialists for each one of these projects. So this one could be using the copywriter for one thing and a different project is calling on the same one. They'll pass that back to the chief. Now, if you just add a bunch of bots and they're all in one flat level, it does not work. Best thing to do is pin this chief bot to the top of your sidebar and then pin active projects. You're not always gonna be using them, but if you're in the middle of something, that lead should be pinned. And then you don't need any of these specialists in your sidebar. You need to be consistent with the shapes and colors, which I'll show you in a second, but that will clean up. 90% of my chats go through my chief. The other 10% through these. I almost never message the specialists. So what you get with your sidebar is something like this, before and after. If you have your chief, which is first, you gotta do that, and then your project leaders up here, you also have all these one-off bots and there's like 15 of them or 28 of them and you can't keep track. What I would do is I would clean that up, keep your pinned chats like this, and then hide all these one-off bots. So you can tell these basically got condensed down into the nine I don't need. And I do have a couple of teams that share leads because they overlap, but for the most part, I message my chief of staff bot and my project leaders. ## Naming, colors, and shapes So how do you organize this with the naming, the colors and the shapes? First of all, use the default designs. Here's what you get for colors when you make a Grok Bot. There's this color palette. You need to use it and be intentional. Don't go make your own. Yeah, they can be personalized, but it defeats the animation, which is like the best part of Grok Bot. They blink at you and the little shapes move around. You can use whatever you want, but this is the color system I use: - My chief is my only purple bot. - My project leaders are unique. So I might have one that's brown, one that's red, and one that's orange. That means you can have a lot of projects going without reusing colors. - Green is for all of the specialists that I don't want to message. They work underneath these leads. So that's my system and it just helps me visually. I don't even have to read the names. The other thing that really helps is the shapes. So what I do is I use three of these and they're all cool, but I stick to three: - The chief, that's one shape and it's the only one that gets that shape. I like the cloud for the chief. - Your leads, I like triangles. - Your specialists, I use squares. I don't use any of these other ones. And you get something like this. That's the leader. I know. I don't need to read this. These are the project managers and they're all different colors, which is related, like this one's with football, so it made sense to make it brown. This one, the podcast is blue, whatever you wanna do. But those are unique and you can rotate through and retire them as needed. But then you can tell by looking at those what they're called and how they fit. So use the shapes and colors. ## Two types of channels You're primarily gonna have two types of channels. One, I call all hands. What that means is my chief of staff is in here running and delegating between all of these different agents. I make them meet every weekday before I'm awake and they'll start messaging each other and it's all about coordinating projects and avoiding conflicts on the calendar. That's your most important channel and where you need to have your chief running all of the other project managers. The second one is you need project channels. So if you make a triangle bot to run this thing, like I have one called Operation Free Agent, you put in, here are the specialists, you get to coordinate between. And I'm still not gonna chat with this one a lot, but what you do see is your chief of staff is gonna pass things to this bot, and then it's gonna say, "Hey, I need the personal brand and the web dev to put up a blog for Aaron and his own style and color scheme or whatever." And those are the only real chats you need. One for all hands, one for each project, which shouldn't be any more than you have triangle bots in this system. Now, people are obsessed with getting templates to bots. I'm gonna tell you, none of mine are any good to anybody. They're like skill files. Use them as inspiration, but copying my skill is not gonna help with your task. I have a [chief of staff](https://x.ai/bot/eiVFbd0nIdH2gzSwHOs0D) bot. You probably have something similar. If you want a copy of it, use the chief of staff template. One that is a single purpose bot, which I stole from some other people and I customize for myself, is an [Overwatch](https://x.ai/bot/HtClSXO_AmiQoyYH9aXV9) bot. And its main job is keep everything organized, backed up in GitHub and portable. So if I hit rate limits, I can take it out. I'm not stuck in Grok Bot or having to pay to get more usage today. Or if Grok Bot's no longer the thing, I don't want all my work to die and have to copy and paste chats. That's silly. So it just has a couple of routines to clean up, organize and back stuff up in GitHub. That one is totally worth stealing if you don't have something similar. And then finally, I made a [Set Up](https://x.ai/bot/BsExflSUXpW0hs21OTBzu) bot for you based on exactly how I set mine up, down to the color and the names. It'll walk you through setting up your own Grok Bot system just like me so yours is no longer a disorganized mess. ## Related pages - [All field notes](/blog) - [Tools and downloads](/tools) - [About Aaron](/about) ## Sources and references - [Watch on YouTube](https://youtu.be/tcG1mSRbNGI) - [Chief of staff template](https://x.ai/bot/eiVFbd0nIdH2gzSwHOs0D) - [Overwatch template](https://x.ai/bot/HtClSXO_AmiQoyYH9aXV9) - [Set Up template](https://x.ai/bot/BsExflSUXpW0hs21OTBzu) --- # I edited a fishing video from the ChatGPT mobile app without touching a timeline Source: https://aaronmakelky.com/blog/codex-hyperframes-video-editing-workflow Last updated: 2026-07-25 The real prompts, screenshots, corrections, and tools behind a 31.7-second vertical video I edited with Codex and HyperFrames from my phone. [Back to blog](/blog) Field notes July 25, 2026 · 12 min read Codex · HyperFrames · AI Video Editing · ChatGPT · Creator Workflows I did not touch a waveform, video app, or keyframe. I gave Codex a Google Drive video, ten fishing photos, and a screenshot of the fishing report it had made for me. I stayed in the ChatGPT mobile app on my iPhone while Codex pulled the files onto my Mac, found my personal video brand system on GitHub, built the edit in HyperFrames, rendered review copies, and sent them back to my phone. The final video is 31.7 seconds. I published it on X, Instagram, and TikTok. [Download the captioned MP4.](/videos/codex-hyperframes-fishing-video.mp4) The finished captioned version. A second render removed the music and captions for native social publishing. [Watch the published post on X.](https://x.com/theaaron/status/2081123137891213576) This was not a one-prompt miracle. The first version missed, and parts of the next several versions did too. The corrections are what made the workflow useful. ## The fishing story behind the video Before I went to the Middle Fork of the Flathead, I asked Codex for a fishing report. It recommended a few dry flies, including a Purple Haze. Instead of checking another report or a hydrograph, I tied on the Purple Haze and caught fish almost every cast. ![Codex fishing report recommending a Purple Haze for the Middle Fork of the Flathead.](/images/blog/codex-hyperframes-video-editing-workflow/codex-fishing-report.jpg) The original fishing report. Purple Haze appears in the dry-fly plan. I had a 52-second vertical clip of me standing in the river talking about AI. Most of it was a longer analogy about what AI can and cannot do. Buried in the middle were the lines I wanted: > "Hear me out, Codex as a fishing report service." > > > > "I have been catching fish left and right based on a fishing report that Codex prepared for me." > > > > "That's like a size 14 Purple Haze." > > > > "But it can't actually go out and cast the fly rod and hook the fish. You still have to go do that yourself." I also had the report screenshot, a photo of the fly on my rod, river photos, and enough fish pictures to make the result pretty difficult to argue with. ![Aaron Makelky in the Middle Fork with the Purple Haze visible on his fly rod.](/images/blog/codex-hyperframes-video-editing-workflow/purple-haze-on-rod.jpg) The same fly from the report, visible on the rod. ## The first prompt I started the task from the ChatGPT mobile app and attached all ten photos. The source video lived in Google Drive. This was the main instruction: > "Here's the media of when I used Codex to help me plan for a fly fishing outing on the Middle Fork of the Flathead. It recommended some dry flies to use, including a Purple Haze, which is in the screenshot." > > > > "I want you to turn that into like a montage-style video, and then show the fly that I used, the fishing report, the fish, etc., as like a funny, kind of tongue-in-cheek style short video for my personal brand." > > > > "Use the Hyperframe skill to do any sort of on-screen graphics, animations, etc., and look for the project folder in Codex that has all of the personal video assets for me. It's in GitHub somewhere." That prompt supplied the story, the evidence, the tone, the format, and the source of the brand rules. It did not specify timecodes, a shot list, animation curves, or caption timing. Codex found those details in the source material. ## What Codex assembled The Drive video was 154 MB, too large for the connector's normal download path. Codex tried the shared link, hit Google's sign-in page, and then used my authenticated Drive session to retrieve the original file without shrinking it. It found my `a-makelky/aaron-video-assets` repository and used the approved brand system at commit `66fd7344245da5e8ddc83340915cdbc51794a2ce`. That repo already defined the visual boundaries: - Navy, blue, white, and teal - Syne for display type and DM Sans for supporting text - Flat color fields and restrained motion - Four-pixel corners instead of rounded template cards - No gradients, particles, bounce, glitch, fake dashboards, or generic AI imagery - One teal accent group per frame - Never cover my face, hands, or the evidence ![Blue, white, and teal field-note graphic from Aaron Makelky's reusable video brand system.](/images/blog/codex-hyperframes-video-editing-workflow/personal-brand-field-note.jpg) One reusable graphic from the video brand system Codex found on GitHub. Codex created a word-level transcript of the source clip and selected three audio sections: | Final scene | Source time | Length | | --- | --- | --- | | "Hear me out" opening | 35.18 seconds | 4.0 seconds | | Fishing report and Purple Haze | 7.52 seconds | 12.6 seconds | | What AI still cannot do | 25.84 seconds | 6.2 seconds | It built the video as five HyperFrames compositions at 1080x1920: 1. Me in the river with the opening line. 2. The report, a tight Purple Haze focus, and the fly on my rod. 3. A rapid sequence of fish photos. 4. Me explaining that Codex still cannot cast the rod. 5. The bent rod and the river with no added conclusion. HyperFrames handled the timeline, graphics, photo choreography, transitions, captions, and final MP4 render. The project pinned HyperFrames `0.7.71` so it can be rendered the same way later. ## The copy was wrong before the edit was right The first draft tried too hard to sound like a social video. It used labels and compressed contrast lines like: > "AI CAN PLAN." > > > > "IT CAN'T CAST." > > > > "CODEX PICKED THE FLY." It sounded like generic social copy, and my response was: > "Trash copy. Drop the field note copy, just do narrative style copy, not this pithy, stacotto copy" That changed the on-screen story to three normal sentences: > "I asked Codex what fly to use on the Middle Fork." > > > > "It told me to tie on a size 14 Purple Haze." > > > > "I tied it on and started catching fish almost every cast." The last two scenes had no story copy. My spoken line and the river were enough. We ran those sentences through my AI-writing audit. The final pass found zero LLM-style issues. ## Mobile review changed the workflow Codex initially gave me a localhost preview. That works when I am sitting at the Mac. I was not. > "I'm on mobile gpt remote, I can't see a localhost video. Share it with me in a medium I can view here" From that point forward, each review came back as a small H.264 MP4 that played in the mobile app. The full-resolution master stayed separate. That mattered because I could pause on a bad frame, take a screenshot on my phone, mark the problem, and send it back into the same task. I never opened the underlying timeline. ## The feedback stayed visual The early edit had a large blue panel over too much of the opening shot. I did not tell Codex which CSS property to change. I said: > "this blue background obscures too much of the frame. move it to the top above my hat, then lets add on brand captions in my brand style for all of the narrated speech" ![Mobile review screenshot showing a blue story panel covering too much of Aaron's fishing video.](/images/blog/codex-hyperframes-video-editing-workflow/review-opening-panel-too-large.jpg) The opening panel covered too much of the original frame. Codex moved the story sentence into the open sky above my hat and built phrase-level captions for the narration. It kept the captions separate from the story copy so the frame did not become one giant block of text. Another scene had a teal line that looked intentional but pointed at nothing useful. > "this scene is poorly framed and the teal highlight obscures text. what is it's purpose? it should be focused in on 'Purple Haze', which is the fly i used the entire time and the one visible on my rod in the still image of me" ![Mobile review screenshot showing a teal line obscuring the wrong part of the fishing report.](/images/blog/codex-hyperframes-video-editing-workflow/review-highlight-wrong-target.jpg) The original teal treatment obscured text without identifying the evidence. The scene was rebuilt around a simple visual argument: 1. Show the actual Codex report. 2. Focus on the exact words "Purple Haze." 3. Cut to the same fly on my rod. 4. Cut to the fish. The teal accent finally had a job. ## "Fix it and check your work" The first attempt at the Purple Haze focus still missed. The box did not fit the words. > "this frame still misses the mark. the teal box doesn't fit the words 'Purple Haze' fix it and check your work" ![Mobile review screenshot showing a teal box that does not fit the full words Purple Haze.](/images/blog/codex-hyperframes-video-editing-workflow/review-purple-haze-box-too-small.jpg) The first Purple Haze box did not contain the complete phrase cleanly. Codex changed the treatment so the box sized itself from the rendered words instead of using a guessed rectangle. A full-resolution check then found another problem: parts of the original screenshot lettering were peeking out behind the replacement. The box padding and position were adjusted again. Codex captured the exact 10.5-second frame and used OCR to confirm where "Purple Haze" sat in the source image. ![Final video frame with a teal box fitting the complete words Purple Haze.](/images/blog/codex-hyperframes-video-editing-workflow/final-purple-haze-frame.jpg) The final focus frame after the full-resolution fit check. This is the part that tends to disappear from AI demos. The useful loop was: 1. Render a review copy. 2. Watch it on the device where I was actually working. 3. Point to the specific miss. 4. Change the project without rebuilding it from scratch. 5. Render and inspect the exact frame again. ## Tightening the audio The first cut waited too long before the opening line. It also held about a second of silence after my final sentence. > "Too much dead air at the start, needs a simple background beat, and the teal highlighted frame doesn't make sense. What is it even highlighting?" And later: > "there is also dead air at the end of my last line, cut it and tighten the edit" Codex used the source waveform to find the last audible word and cut the shot 0.13 seconds after "yourself." The river ending starts immediately after it. The runtime dropped from 33.5 seconds to 31.7 seconds. ## The music route failed I asked Codex to use MiniMax CLI to generate a simple background beat. The CLI was installed and authenticated, but the account returned "no active Token Plan" or "usage limit reached" for every available music model. There was no reason to keep retrying the same account. I had a licensed track from my paid Melodie account, so I sent the local file path: > "minimax sub is dead apparently, nevermind. find a different way, here's a song i downloaded from melodie, which i have a paid sub for '/Users/aaronmakelky/Downloads/MEL871_03_1_Cowgirls_Dont_Wait_(Full)_Enrico_Liverani.mp3'" Codex cut a 31.7-second cue from "Cowgirls Don't Wait" by Enrico Liverani. It mixed the track quietly under my voice, raised it during the silent fish sequence, ducked it before the final spoken line, and faded it under the last river image. The generated-music tool failed, so we used the licensed music I already owned. ## How the video was checked The final review went beyond confirming that the render completed. HyperFrames checked the project for runtime errors, broken layout, motion problems, safe-area collisions, and contrast issues at eight points across the timeline. Codex also generated contact sheets and inspected the important frames at full resolution: - Opening story copy above my hat - Full fishing report - Purple Haze focus at 10.5 seconds - Purple Haze visible on the rod - Fish sequence - Caption placement during both talking-head sections - Final river frame with no copy The full master was rendered at 1080x1920. A smaller 720x1280 H.264 copy was made for review in the ChatGPT mobile app. FFmpeg handled the mobile encoding, and FFprobe verified the duration, dimensions, H.264 video, and AAC audio. ## One project, two delivery versions After publishing the captioned version on X, I wanted a second file for Instagram and TikTok with no background music and no narration captions. I planned to add those inside the social apps. > "I need you to make a duplicate version of the video, but no background music and no captions. I will add those in natively on some video social media apps." Codex added a reusable `nativeSocial` export switch to the HyperFrames project. It did not delete anything from the original. The normal render keeps the licensed music and branded captions. The native-social render removes both while preserving my voice, river audio, story graphics, and timing. Codex verified that speech remained audible and that the fish montage and closing section were digitally silent where the music had been removed. When I said I needed the full-resolution file on my iPhone, Codex uploaded the 1080x1920 master to Google Drive and sent the download link back into the chat. ## The reusable version This is the starting prompt I would use again: ``` Here is one source video and a folder of supporting screenshots and photos. Build a vertical social video from the strongest spoken lines in the source. Use the real footage as the narrative spine and treat the screenshots and photos as evidence, not decoration. Use my canonical personal video brand system. Use HyperFrames for the edit, on-screen graphics, captions, transitions, photo timing, and render. Keep the source voice and ambient sound. Do not generate a voice or use generic B-roll. Use simple narrative copy in first person. No labels, slogans, fake dashboards, or retention-editing effects. Create a lightweight review MP4 I can play in the ChatGPT mobile app. After each revision, inspect the exact frames I flag and run the full HyperFrames check before rendering again. Deliver: 1. A 1080x1920 master with branded captions and licensed music. 2. A second 1080x1920 version with no music and no captions for native social publishing. 3. Mobile-friendly review copies. 4. A project brief, storyboard, transcript, and final delivery notes. ``` The final thing I asked for was not another creative revision. It was a full-resolution version I could download on my phone, then a clean duplicate for native captions and music. Codex made both and put the master in Drive. ## Related pages - [All field notes](/blog) - [Tools and downloads](/tools) - [About Aaron](/about) ## Sources and references - [Published video on X](https://x.com/theaaron/status/2081123137891213576) - [HyperFrames](https://hyperframes.heygen.com/) - [OpenAI Codex](https://github.com/openai/codex) ## Common questions ### Did Aaron edit the video in a traditional video editor? No. Aaron reviewed the video and sent corrections through the ChatGPT mobile app. Codex changed and rendered the HyperFrames project on his Mac. ### What did HyperFrames do in this workflow? HyperFrames handled the code-based timeline, scene composition, on-screen graphics, captions, photo timing, transitions, checks, and final MP4 render. ### Can this workflow create a version for native social captions? Yes. The project includes a native-social export that removes the music and narration captions while preserving Aaron's voice, river audio, story graphics, and timing. --- # He told me he felt like he could finally read Source: https://aaronmakelky.com/blog/ai-in-education-friction-question-not-tool-question Last updated: 2026-03-27 A student told me the tool let him read. The hard part was deciding which friction was still the work. From a Skilling Future Podcast conversation about AI inside a school. [Back to blog](/blog) Field notes March 27, 2026 · 10 min read Classroom AI · Literacy · Podcast A student in my high school social studies class read at a third-grade level. One of the hardest-working kids I have ever taught. We handed students like him college-level readings and said good luck. So I taught him to build a system in ChatGPT around how he actually learned. He liked working out. He liked hunting, fishing, and trucks. He knew what reading level felt manageable. He could take a speech from a 19th century German chancellor and get it back in language that clicked for him, with analogies that connected to his world, not mine. After a few weeks he told me it felt like a step stool that put him on the same shelf as everyone else in the room. He could sit in the back row with his keyboard and get himself unstuck without needing me to hover over him every five minutes. Nobody taught him that. Nobody gave him permission. He just needed someone to show him the tool existed and let him make it his own. I talked about this with Andy Lucchesi on the [Skilling Future Podcast](https://www.youtube.com/watch?v=JoZXbV-w8Bo), and the conversation kept circling back to one idea: AI removes friction. The hard part is deciding which friction helps students learn and which friction just gets in the way. ## I was bad at using it I first saw ChatGPT the way a lot of teachers did. A student shoved a phone in my face in November 2022 and said, hey, you're into technology, have you seen this thing? I had not. I went home that night, signed up with a Google account, and started messing with it. I did not understand what it was or how it worked. But I had the same gut reaction a lot of classroom teachers had early on: this is going to change how we teach and how students learn. I just did not know how. So in January 2023 I did the most teacher thing imaginable. I tested it on my own assignments. I wrote the essays, took the quizzes, ran the projects through ChatGPT before my students did. At first the output was bad enough that I thought, all right, this thing is not replacing me anytime soon. Then came the more uncomfortable realization: I was bad at using it. The results improved as I got better at asking. I had already lived through one version of this as a history teacher. I started in a world of textbooks, paper notes, and waiting around for information. Then came Google, Wikipedia, and a laptop in every student's hands. A lot of friction around just finding stuff disappeared overnight. I was one of the history teachers who decided memorizing facts could not be the whole point. It is part of learning, sure, but it is not where you should finish. AI felt like the next version of that same shift. A personalized Wikipedia that can also help you study, summarize, translate, and reorganize information around the way you actually learn. ## If the essay takes 10 minutes now, what do we do with the other 80? Writing is a huge part of secondary social studies. Traditionally that means writing workshops, conference time, tutor centers, and one teacher trying to help 30 kids at once. There is only one of me. There are always more students who need feedback than there is time in the period. So I started asking a basic question: if I can write clear enough instructions for a custom GPT to help students brainstorm and revise before they submit, why would I not do that? If a student wants writing help at 2:00 AM on their phone, great. That removes friction from the feedback process. I do not have to personally do all of that anymore. What scares a lot of educators is what comes after. That essay used to take 90 minutes in class. Now a student can do it better in 10 or 15. So what do we do with the rest of the time? I think you use that time for the stuff that needs a teacher in the room: discussion, revision, argument, and feedback. If AI handles blank-page panic and first-pass organization, class time can move toward the work students cannot get from a blank chat box. One of the simplest things I started doing was telling students: here is the assignment we have always done, now take Gemini canvas mode and build something that shows me what you learned. I told them I was not grading the end product. I was grading the thinking process I could see in the chat log. Pretty soon I had students building Oregon Trail style decision-making games. One student coded a history game and made me play it as the assessment. I had to move through historical events as if I was the leader, and the game told me whether I was right or wrong. He made me play the game he built to prove he knew the content. I never would have thought of that. But when you give students a blank canvas and say go, you find out what happens when they stop treating the essay as the only serious container for thought. ## Just use what you already have A huge struggle for educators is IT departments blocking things. It takes six months (or longer!) to get a site unblocked. My advice is always the same: you probably already have access to something. Just start with what you have and get good at it. Last school year, nobody in my building knew they had Gemini. Google baked it into the apps for education suite. No memo went out, staff and students had it on their accounts but nobody told them. People love to ask me what the best model is, what the best app is. I tell them to use the one they have. Until you find things it cannot handle, why are you looking for something else? If you are a Microsoft school, you have Copilot. If you are a Google school, you have Gemini. They are not always the best tools on the market, but they are the most accessible. You do not have to waste time evaluating vendors and finding the budget. Start with what is already open and already paid for. ## Banning AI is training kids for a world that no longer exists I understand the fear. A lot of resistance comes from a good place. Teachers want students to think, struggle, write, and learn on their own. I want the same thing. Where I disagree with a lot of the response is in the solution. There is a faction in every faculty that wants to block the tools and go back to paper and pencil. My challenge to those people: are you serving students for the future they are going into, or are you trying to prepare them for 1988 when you went to college? No boss is going to tell a graduate to write reports by hand so they know the employee did not use AI. They are going to say use it responsibly and turn in something good. That is the standard everywhere outside a classroom. Meanwhile, inside school, we are mostly using Chromebooks to do digital worksheets. A screen version of what we did with textbooks in the nineties. I also do not think forcing every student to use AI is the answer. I put an opt-out clause in my syllabus. If a student refuses to use generative AI, fine. They get the same assignment held to the same standard with a different tool. I use this analogy: if I am teaching you to build a house with power tools and you object to electricity, grab a hammer and a hand saw. I still expect the house built to spec. I have had three students in the last two semesters say they refuse to use AI. My answer is always the same. You can opt out and do the same work a different way. The standard does not drop. I am not going to shame anyone into adoption. But I am also not going to pretend the world outside school is waiting for them to catch up. ## The biggest blind spot is data One of the dilemmas that keeps me up is how little people understand about what happens to their data inside these models. People have an email mindset. They think data works like a message you accidentally sent to the wrong person. Just hit delete and it is gone. Large language models do not work that way. Take a special educator writing an IEP, a very sensitive document with personally identifiable information everywhere. They go home, open ChatGPT on a personal device on the free version, paste the whole thing in. Their boss would never know. But that data is now training the model. You agreed to that when you signed up. And you cannot go to OpenAI headquarters, grab a server, and unplug your information. A lot of my students were using ChatGPT in a phone browser, not logged in, no app. I told them I understand why. It is low friction. But they have the least privacy protection possible and zero control over their data. Names, dates, and their own ideas go in with no way to pull them back out. Schools that teach AI literacy but skip data literacy are setting kids up to get burned. ## Scaling AI in schools is a leadership job I did not start by asking my school for a microphone or a training slot. I just put my ideas out on LinkedIn, TikTok, YouTube. Everywhere except inside my own building. I wanted to see what stuck and whether I had any idea what I was talking about. Instead of targeting my colleagues, I let the work speak for itself. Eventually they came to me. Could you do some training? Could you present at a staff meeting? Sure. Most schools already have what a friend of mine calls AI scouts. They already went through the messy part, hit the dead ends, and figured out which objections are real and which ones are just fear wearing a nicer outfit. If you are a school leader, that is your shortcut. Let your scouts absorb the trial and error, then put them in front of everyone who is three or five or ten steps behind. For every objection a resistant staff member raises, take that list to your scouts. They will know the answer. Then you have to have the courage to say this is how it is going to be. Scaling AI in schools is barely a technology problem. It takes someone willing to have the hard conversation and tell people that ignoring this is not an option here. ## You do not need all the answers You just have to be willing to examine your practice honestly and admit there might be a better way. The thing that connects all the AI scouts I have met, Andy included, is they can look back at how they taught two years ago and say that is not the best way anymore. We live in different countries and teach different subjects, some of us in different languages. But the through line is the same: we looked at what we know now and admitted we needed to change how we work. If you can get past the ego hit, the path gets simpler. Most of the information is free. The harder part is finding people who have already done the reps and being willing to say, hey, maybe there is a better way to do this. ### Where to find Andy Lucchesi and Skilling Future - [Skilling Future Podcast episode](https://www.youtube.com/watch?v=JoZXbV-w8Bo) - [Skilling Future website](https://skillingfutureai.com/) - [Skilling Future on LinkedIn](https://pk.linkedin.com/company/skilliing-future) - [Andy on LinkedIn](https://www.linkedin.com/in/andys-everyday-english/) - [Andy on Instagram](https://www.instagram.com/andys.everyday.english/) ## Related pages - [All AI notes](/blog) - [Tools](/tools) - [About Aaron](/about) ## Sources and references - [Descript](https://www.descript.com/) - [Perplexity](https://www.perplexity.ai/) --- # I cleaned up ten skill folders so you can see the wiring Source: https://aaronmakelky.com/blog/10-openclaw-skills-running-on-raspberry-pi Last updated: 2026-04-07 Hardware is a Pi 5. Each folder has a SKILL.md, a SETUP.md, and scripts only where a prompt is not enough. [Back to blog](/blog) Field notes April 7, 2026 · 8 min read OpenClaw · Raspberry Pi · Agents I cleaned up ten OpenClaw skill folders from my Raspberry Pi setup so you can inspect how each job is wired. Some are scheduled jobs. Some are reference folders. I mostly wanted the folder structure visible: a SKILL.md, setup notes, and scripts where the workflow needs them. ## The setup Hardware: Raspberry Pi 5, 8GB RAM, running Raspberry Pi OS. OpenClaw handles the agent layer. Each folder has a SKILL.md for the job, a SETUP.md for the boring install parts, and scripts where the workflow needs actual code instead of another prompt. ## The 10 skills ### 1. Morning Briefing Delivers a daily briefing to Telegram at 6:30am with weather, calendar events, market data, AI news, and personal notes. Pulls from Google Calendar API, wttr.in, and web search. Python, Google Calendar API, Brave Search, Telegram ### 2. Daily Video Learning Builds transcript-backed YouTube learning packets and saves them where they can be reviewed later. YouTube Data API, Supadata API, OpenAI, Notion ### 3. Weekly Review Turns the week into a short review of content output, project status, and what needs attention next. Bash, files, weekly notes ### 4. Nightly YOLO Overnight build-and-research slot that picks one real problem and ships a concrete artifact, then records what not to repeat. OpenClaw agent, memory files, research log ### 5. Content Flywheel Takes raw notes from the intake stage, drafts social posts in a consistent voice, audits for AI patterns, generates hooks, and queues them for review. Python, LLM drafting and auditing, File pipeline ### 6. Geopolitical Briefing Weekday briefing on war, defense, and cybersecurity, with sources separated from the final summary. Bash, Brave Search, notes ### 7. Social Media Intake Pulls transcripts and content from YouTube, Instagram, and TikTok URLs. Used by other skills for video learning and content intake. Supadata API, Python ### 8. Weather Pulls current conditions and forecasts from simple weather sources so other jobs do not need to solve that part again. wttr.in, Open-Meteo ### 9. Recipe Processor Turns recipe-card images into structured recipe entries with OCR, cleanup, and Notion output. OpenAI, Notion, OCR workflow ### 10. Google Calendar Manages Google Calendar through OAuth so calendar work can happen through an actual API instead of copy-paste. Google Calendar API, OAuth2, Python ## What makes these different from prompt collections Most OpenClaw "skills" shared online are prompt files: a SKILL.md with instructions and nothing else. These folders show more of the wiring: - Actual scripts: Python and Bash files that run API calls, parse data, and write output - Schedule notes: where a job is meant to run on a schedule, the timing is written down - Setup guides: SETUP.md files for API keys, dependencies, and configuration - Known edges: enough setup detail to see where the folder will need your own keys, paths, and judgment The difference matters when you are setting up a morning briefing and the calendar API integration does not work because the token path is wrong. Or when your content pipeline fails because the prompt assumes a folder name you do not use. A folder with setup notes gives you something to inspect before you start changing your own stack. ## Example schedule from my setup ``` 6:00am - Morning briefing 2:00pm - Daily video learning 3:00pm - Content flywheel weekday - Geopolitical briefing Monday - Weekly review 10:00pm - Nightly YOLO build ``` ## Why a Raspberry Pi The Pi runs 24/7 at ~3-5 watts, with no cloud costs or server management and no VM getting patched and rebooted under you. OpenClaw runs as a systemd service and auto-restarts on failure. The whole setup costs $80 in hardware and $0/month to operate. I chose this over a VPS because the agent runs cron jobs that need reliable timing and persistent state. A $5/month VPS works too, but the Pi gives me physical control and zero recurring cost. Get the 10 folders with setup guides and scripts where the workflow needs them. [See the OpenClaw kits I'm selling now](/shop) ## Free resources - [Reliability notes](/reliability): 7 habits for keeping small agents from failing quietly - [AI workflow checklist](/checklist): a quick check before a workflow gets trusted - [Agent setup check](/scorecard): 5 questions about identity, memory, scheduling, and recovery Questions? Find me on Telegram at [@aaronmakelky](https://t.me/aaronmakelky) or at [aaronmakelky.com](/). ## Related pages - [OpenClaw kits](/shop) - [Reliability notes](/reliability) - [Tools](/tools) ## Sources and references - [Raspberry Pi 5](https://www.raspberrypi.com/products/raspberry-pi-5/) - [OpenAI Codex GitHub repository](https://github.com/openai/codex) --- # most teacher AI apps are wrappers Source: https://aaronmakelky.com/blog/3_ai_apps_teachers Last updated: 2025-03-03 Most teacher AI apps are wrappers around the same models. I would start with the frontier tools, then add classroom-specific apps only when they save real time. [Back to blog](/blog) Field notes March 3, 2025 · 12 min read Classroom AI · Tools Teachers do not need another app shelf. Most AI tools sold to schools are wrappers around the same models, and you do not have hours to test every shiny classroom add-on just to find out it saves you five minutes once. I would start with the frontier models, the general-purpose AI tools that power most niche teacher apps anyway. Then I would add classroom-specific tools only when they solve a real problem better than the base model. Video version: [YouTube](https://youtu.be/4nK-cWe_EWI) Audio version: [Podcast](https://share.transistor.fm/s/73261327) ## Why frontier models matter Before we get into the tools, a quick reminder: skills beat wrappers. Prompting, iterating, and checking sources transfer across almost every AI tool you will try. Once those habits are solid, the app choice gets much less dramatic. Even with those skills, picking the right AI tool matters. For teachers, that means focusing on the frontier models. For teachers, I would start with these three: ### ChatGPT: the all-around default You've probably used it. It's the most common AI, and for good reason. Its first draft often sounds generic. Projects and saved context help, but I still have to steer it. #### Where it works - Everyone knows it, including your students. - OpenAI keeps adding more features to the free version. - Advanced voice mode is great for real conversations and translation. - The Windows and Mac apps let you skip the browser. The mobile apps are good too. #### Where it gets annoying - Even with the "search the internet" feature, it is not the search tool I reach for first. - Honestly, the interface looks like engineers made it, not designers. - The model names are long and confusing. I would start with ChatGPT. It's a solid all-around tool and you probably already have it. ### Google Gemini: the Google school pick This one might be surprising. Early versions of Google's AI, back when everyone still called it Bard, were rough. Compared with those days, Gemini became the one I kept reaching for. #### Where it works - The writing is more natural and less "AI" than ChatGPT's first try. - It handles images and different document types. - If your school uses Google Workspace for Education, this is a must. Gemini is getting built into Docs, Gmail, Sheets, and Calendar. - The voice mode is second only to ChatGPT. #### Where it gets annoying - You need to pay for Gems, Google's version of custom GPTs. They also are not shareable with others, unlike Custom GPTs. - The "Deep Research" tool is cool but still needs work. - Many people still remember the early, bad versions. If you're in a Google school, Gemini belongs on the shortlist. Even if you're not, test it against what you already do in Google Workspace. The integration is the reason. ### Perplexity: the search tool I would test first Perplexity became the AI search tool I kept using after the test. I replaced Google Search with Perplexity in all my browsers and watched what happened to my normal search habits. #### Where it works - A paid Perplexity plan lets you compare multiple models inside one search interface instead of hopping between apps. - The interface puts sources up front and keeps the page clean. - The desktop and mobile apps are good, including searching from a picture on mobile. #### Where it gets annoying - Most teachers haven't heard of it. - The free version is made for speed, not writing. I would change your default search engine to Perplexity for a week and see what happens to your normal search habits. ## Two I would re-test before recommending This was my early-2025 read. I would re-test this section before treating it like a current ranking: - Claude: Some people love it, especially coders and writers. I still would not start most teachers there unless they need artifacts or already like its workflow. The first three are easier classroom starting points. - DeepSeek: DeepSeek is free and strong at reasoning, but there are problems. It shows its "thinking", but there are privacy concerns because the data goes through China, plus a strong pro-China bias if you ask about Taiwan. More importantly, Perplexity already has a US-hosted version of DeepSeek's R1 model without the same classroom privacy problem. ## Where I would start If I were advising a teacher in 2025, I would put the first week into ChatGPT, Gemini, and Perplexity. Claude and DeepSeek can wait: Claude is better for other lanes, and DeepSeek has too many classroom/privacy problems unless you are using it through something like Perplexity. Try these apps in your actual work. Reading about them will not tell you much. If you try one of these in a classroom and hit a weird edge case, send it over. That is usually where the useful part starts. ## Related pages - [All AI notes](/blog) - [Tools](/tools) - [About Aaron](/about) ## Sources and references - [Descript](https://www.descript.com/) - [Perplexity](https://www.perplexity.ai/) --- # revision history may not catch the agent Source: https://aaronmakelky.com/blog/agentic-ai-classrooms-what-teachers-need-to-know Last updated: 2026-02-20 Agentic browsers can move through web assignments in ways revision history may not catch. That changes what a teacher can verify after the work is done. [Back to blog](/blog) Field notes February 20, 2026 · 12 min read Classroom AI · Agentic AI I hit "run" on an agentic browser, handed it a Google Doc full of questions about the League of Nations, and walked down the hall to get a drink of water. When I looked through the classroom window, my high school students were staring at the screen while tabs opened and answers typed themselves into the document. A few of them thought I was remote-controlling the computer. I just waved. Watching that, I stopped thinking of AI as a copy-paste shortcut. It could work on its own now. I sat down with Lindy Hockenbary on her [Make EdTech 100](https://makeedtech100.transistor.fm/episodes/education-in-the-ai-agent-era-with-aaron-makelky) podcast right after TCA conference in February 2026 to talk through the latest round of AI releases. We came in with a list of updates and left rethinking what these tools actually mean for classrooms. Prefer to listen? [Catch the full episode with Lindy here.](https://makeedtech100.transistor.fm/episodes/education-in-the-ai-agent-era-with-aaron-makelky) ## A packed room of educators had never heard the word "agentic" At TCA, Lindy was sitting in a session when the presenter mentioned agentic browsers. The room, packed and standing room only, went sideways. Most people had never heard the word "agentic." Some who had heard it immediately started asking how to block it. Agentic AI has been the main thread behind most of the big releases over the past year, yet the majority of educators at a major tech conference were encountering the concept for the first time. Here's the simplest way I can explain it: agentic AI has tools. If you gave ChatGPT a toolkit, a wrench for this job and a hammer for that one, that's what makes it an agent. Maybe it gets a browser and can run a search. Maybe it can write to your Google Doc. Maybe it fires on a schedule. Monday mornings at 8 AM, run this task. Maybe it's a trigger: when I get an email from this person, do this. The brain is the same AI you've been using. The tools make it an agent. Generative AI waits for you to prompt it, copy the output, and paste it somewhere. Agentic AI goes and does the thing. You tell it to respond to your emails saying you're out next week at NCC in Seattle, and it opens your inbox, drafts the replies, and hits send. The autonomy is what freaked people out in that session. ## Revision history is not a safety net anymore If you're an educator relying on anything in a web browser to assess student learning or track progress, you need to hear this: a browser agent can do the assignment for the student, and the revision history will look organic. One way teachers have been catching AI-generated work is checking the revision history of a Google Doc. If a student pasted in a four-page essay all at once, that was the tell. That worked for about a year. It doesn't work now. A browser agent doesn't paste one giant block from a ChatGPT tab. It navigates into Canvas or Schoology or Google Classroom. It shows the student's account logged in. It types at a realistic speed. I've had students show me that you can prompt it to add a couple of typos, then go back and correct them. The whole thing runs without the student touching the keyboard, but the revision history looks like a person sat there and wrote it. There are even Chrome extensions built for this. You paste your AI-generated essay in, set the typing speed, and it goes into your Google Doc character by character. Those extensions are free. They're on the internet right now. Not every student is doing this, but the option exists, and the students who want to find it will find it on TikTok or YouTube Shorts. They're not going to see videos about responsible AI use in their feed. They're going to see "how to make your Google Doc look like you wrote it yourself." That's one of the reasons I'm a big believer that educators have to be the ones introducing these tools to students. You can't win the blocking game. Your administration decides what goes into your Google Apps for Ed account, and they might listen to your input, but probably not. Or a committee takes three months to decide. In the meantime, you're stuck with whatever's available, and you have to make it work. ## Three layers of AI tools I would actually try Lindy asked me to walk through the tools, and I think a layered approach makes the most sense. The first layer is the agents built into ChatGPT and Claude. These have been around for over a year. In ChatGPT, you hit the tools button, turn on agents, and ask it what it can do for you. If you've been using ChatGPT regularly with the memory setting on, it already knows your context. It knows you teach high school social studies or that you run a department. You don't have to give it background. Just ask it for customized suggestions and go from there. The second layer is Gemini built into Chrome. If you use Chrome, and most K-12 educators do, there's now a Gemini button in the top right corner. You click it, and it can see whatever tabs you have open. No more downloading a Google Doc, uploading it to ChatGPT, and copy-pasting links. Gemini already has the context of your open tabs. For teachers, that removes a few annoying steps. For students, it removes the same friction that was supposed to slow them down enough to think. The third layer is standalone agentic browsers. This is where Comet, Atlas, and Dia live. What Google and Claude did was put extensions into Chrome and let those control the browser. Perplexity took a different approach with Comet. They forked Chrome's architecture, kept your Chrome extensions working, and rebuilt the whole thing with the agent baked in from the start. I tested Comet with the latest Opus 4.6 model, and it ran a task autonomously for about 30 minutes, taking somewhere between 175 and 190 actions as it scrolled, clicked, typed, and navigated. It's not faster than doing it yourself. Maybe half the speed. The useful part is that you set the task, let it run in a tab group, and go do something else. Twenty minutes later, it's done or it tells you where it got stuck. Atlas does the same thing but uses your ChatGPT account, so it's better if you're already invested in that ecosystem. Dia browser is another one worth looking at. Same company that built Arc browser, which is still probably my favorite browser of all time. They made an AI-native version of it. Free tier available, but the good models require a subscription. ## Codex can help with folder work too When OpenAI released Codex in early February, I spent an entire weekend in it, not because I had to, but because it clicked. A lot of the vibe coding and AI coding apps haven't felt relevant to teachers. What would an English teacher do with a coding tool? Codex changes that. It's ChatGPT's model built specifically for code, running in a standalone app with local file access. You don't need a terminal. You don't need to know what a terminal is. Here's what that means in practice. I do some analytics reporting. I can download CSV files, put them in a local folder, open Codex, and say "go analyze everything in this folder." Before, I'd be uploading and downloading files through a browser tab. Now it just reads what's on my machine. I took a website that originally took me 10 to 12 hours to build over multiple revisions. I told Codex to redo the whole thing, and it rebuilt the site with a new design, new content, and a different domain in about thirty minutes. Then I thought, okay, it's good at design. What's a design task that a non-technical person would actually do? So I had it make my LinkedIn banner. It produced something better than what I could do in Canva or Figma, in about a minute. And it saved directly to a folder on my machine without exporting artifacts or downloading PDFs. I just said "put it in this folder and call it designs." OpenAI publicly said they used their newest model to build Codex. The tool was already being used to build the next version of itself. ## OpenClaw is an empty house without utilities The saga of ClaudeBot to MoltBook to OpenClaw made for good social media content, but the actual tool is more limited than the hype suggests. OpenClaw doesn't do anything new. You could have duct-taped your own version together with Claude Code or other tools if you were technical enough. What's different is the interface. It made that kind of setup accessible to regular people. You can message your agent through WhatsApp, and it can search the internet or reach into your cloud files and calendar. But the app is free in the same way an empty house is free. You still need to connect it to the power grid. It needs an AI brain, a ChatGPT or Claude API key, and those cost money to run. The Best Buy in San Francisco ran out of Mac Minis because people were buying dedicated machines to keep OpenClaw running 24/7. Most of those people are going to figure out pretty quickly that the tool is free but the services cost real money. My honest take: you'll spend more time trying to make OpenClaw work than you'll find useful things for it to do. The one real advantage is making AI accessible to people in your life who aren't technical. I set up a version where my wife can text a question about our family calendar, where's the next wrestling meet, what time is student council in the morning, and get an answer in plain language through Telegram. That one job is the useful part. And whatever you do, run it on a separate device. Don't give it access to your work computer, your school email, or anything with sensitive information. Think of it this way: you would not give a student teacher your bank account password and your personal laptop 24 hours a day. Don't give that to an agent either. ## Where this leaves educators Lindy made a comment near the end of our conversation that I keep coming back to. Formal education isn't going to stay relevant if it doesn't adapt. I agree with that, and I'd add a qualifier: adaptation means understanding what these tools can do, not adopting every one of them, so you're not blindsided when a packed room full of educators encounters the word "agentic" for the first time. The students are going to find these tools whether we introduce them or not. They already have. We can show them what responsible use looks like, or we can leave that job to their TikTok algorithm. I don't have a perfect answer. But I'd rather be the teacher who walked down the hall and let my students watch the browser work on its own, came back, and had an honest conversation about what that means, than the one who pretended it wasn't happening. ### Where to find Lindy Hockenbary - [Make EdTech 100 Podcast](https://makeedtech100.transistor.fm/episodes/education-in-the-ai-agent-era-with-aaron-makelky) - [LinkedIn](https://www.linkedin.com/in/lindyhockenbary/) - [Instagram](https://www.instagram.com/lindyhoc/) - [YouTube](https://www.youtube.com/@LindyHoc) ## Related pages - [All AI notes](/blog) - [Tools](/tools) - [About Aaron](/about) ## Sources and references - [Descript](https://www.descript.com/) - [Perplexity](https://www.perplexity.ai/) --- # cheaper tools did not make anyone sound like themselves Source: https://aaronmakelky.com/blog/podcasting-for-creators-human-content-wins-2026 Last updated: 2026-02-20 Cheaper tools did not fix the real problem: creators still have to sound like themselves. Format choices and a few simple systems keep a creator from sounding interchangeable. [Back to blog](/blog) Field notes February 20, 2026 · 10 min read Podcasting · Creators I think the easiest way to fail in podcasting is to assume access is still the hard part. It is not: I can record on a phone, upload in minutes, and put something on Spotify before most people have finished setting up their mic stand. The real challenge now is bringing something useful that a hundred other creators cannot copy. I joined Dr. Alfonso "Fonz" Mendoza on My EdTech Life to talk about content creation, podcasting, and what has changed since AI tools became normal for creators. The short version: cheaper tools did not fix the trust problem. Prefer to listen? [Catch the full My EdTech Life episode here.](https://www.myedtech.life/makelky/) ## Starting is easy now, so standing out is the hard part When I started, I did not start for sponsorships. I started because I was trying to convince a friend to make a podcast, and he said editing was too much work. So I made my own just to learn, with no audience plan and no monetization plan, just reps. Most people overestimate what they need to begin and underestimate how much they can learn by publishing quickly. One of my students once told me to download a free app and just record on my phone. I talked into my iPhone and uploaded. It was not polished, but it was live in about ten minutes, and that reset how I think about the medium. If someone has an opinion worth sharing, or guests worth learning from, I think they can start now instead of waiting to buy a better camera or design the perfect intro. The issue is that this same access is available to everyone else. So yes, entry is easy. But if entry is easy, differentiation matters more. You do not need a stunt, just a reason for people to keep listening. That reason can be a clear audience, a tighter format, a direct teaching style without fluff, or simply a promise you actually keep every week. When Dr. Fonz asked about the shift in podcast potential, that was my answer underneath the tools talk. More people can create, which is great. It also means creators need a sharper point of view. ## Stop chasing big download numbers and build the right relationships I still see creators get stuck on vanity metrics. How many downloads did I get, how many subscribers, and why do bigger brands not want to sponsor me yet? I understand the frustration, but I think it leads new creators into the wrong strategy. If you launched three weeks ago and have fifty downloads, large brand deals are unlikely. Your first phase is learning and trust-building. I usually tell people to "sponsor yourself" first. Keep creating until your data is useful and your audience profile is clear. In parallel, join network-style opportunities where you can learn how sponsorships are run, how ad reads are placed, and how results are measured. Even if the money is tiny at first, that experience is not tiny. I also think new creators underestimate guest strategy. If you are a startup or a solo consultant, bring in the people you want to work with. Have strong, specific interviews with them. Publish publicly as a byproduct. Your real value can come from the relationships you build before you ever close a "real" sponsor. When Dr. Fonz pushed on whether sponsorship culture is changing, I gave the same answer I give in private calls: big numbers are less meaningful than they used to be, and relevance is more meaningful than it used to be. Ten thousand vague listeners may feel good, but a thousand right-fit listeners can change your business. ## Format changes whether people actually get to it I have seen great ideas disappear in bad packaging. I have seen this in schools, agencies, and creator businesses. Teams spend hours writing long text updates that people never read. The same information in audio or short video gets consumed immediately. That happens because people are busy, not lazy. I think format is about context. My wife is a good example. She listens while doing other things around the house. She is not opening a long PDF for leisure. A lot of people consume learning content the same way. This is why I push creators to capture audio and video, even if they think they only want one output. Record once, then repurpose. Use the long form for YouTube and podcast platforms. Pull short clips for social posts. Turn strong points into text if your audience likes written summaries. One source can support every channel without reinventing your content each time. I also disagree with the "audio-only forever" stance for most new creators trying to grow. If you want to stay audio only for creative reasons, that is valid. I respect that. If the goal is discoverability, it is harder now without video distribution. Platforms reward video behavior. Ignoring that reality is a strategic choice, not a neutral one. Dr. Fonz and I both came from education, so this point matters even more there. Teachers and schools do not need to become media companies. They just need a repeatable publishing habit that meets people where they already are. ## Repeatability is where speed comes from The first time I edit, everything feels slow. By the tenth time, I move faster if I build repeatability into my process. This is why I care less about "best tool" debates and more about whether a creator can produce consistently without burning out. A practical benchmark I use is a one-to-one ratio between final runtime and total edit time. If I record for one hour, can I move from raw file to publish-ready in about one hour? New creators will not hit that immediately, and that is fine. But it is a useful target because it forces better decisions. Templates, saved layouts, reusable intros and outros, clip presets, naming conventions: anything you set up once and stop rebuilding. Anything you can reuse should be reused. I have seen creators waste huge amounts of energy rebuilding the same visual setup every week. Then they wonder why publishing feels impossible. This is where AI can be practical without taking over the creative work. For example, I like using AI to identify potential clips, place markers, or clean caption punctuation. I still review every output before posting. The goal is to remove repetitive grunt work, not outsource judgment. If a system saves me twenty minutes on low-level editing, I can spend that time improving the actual message. ## I trust a rough edge more than over-editing There was a period where everyone chased maximal editing: heavy transitions, constant effects, aggressive cuts, every filler word removed. A lot of it looked impressive and felt exhausting. I think the trend is moving back toward clarity and human voice. When I edit, I usually leave natural filler in audio and video unless it muddies the point. For captions, I often clean it up so the text is easier to read. That balance keeps the person in the delivery while making the idea easier to follow. Over-editing can do more damage than light imperfection. If cuts are too aggressive, people feel the jump even when they cannot explain why. That friction pulls attention away from the idea itself. I have watched highly respected shows where the visual cutting style was so tight I had to switch to audio-only consumption. Perceived quality is not the same as over-processing. The edits I trust are the ones that make the point easier to follow without sanding off the person saying it. This applies to AI use too. I use AI every day and work in a space where these tools are central. What I am against is outsourcing all thinking. When I skip AI entirely for a day and do things manually, it reminds me where my own weak points are. That check matters. If I automate every step for too long, I feel the cognitive drop. Most creators will if they are honest. For podcasting, I trust AI most when it is doing support work, not trying to be the creator. I want tools speeding up production without flattening the judgment that made the show worth recording. ## What I still trust I think podcasting still works because relationships still work. People share things made by people they trust. They send links to friends. They talk about ideas in group chats. They come back when they feel like someone is speaking directly to their reality. I do not see avatar trends or automation stacks replacing that. The creators I trust keep showing up with a clear point of view and something specific to say, no complicated setup required. Trust is still the hard part. ### Listen to the episode - [My EdTech Life with Dr. Alfonso Mendoza](https://www.myedtech.life/makelky/) ## Related pages - [All AI notes](/blog) - [Tools](/tools) - [About Aaron](/about) ## Sources and references - [Descript](https://www.descript.com/) - [Perplexity](https://www.perplexity.ai/) --- # context, a revision, then check the work Source: https://aaronmakelky.com/blog/ai_skills_everyone_2025 Last updated: 2025-02-22 Before the app choice matters, students and teachers need a few repeatable habits: give the model context, push for a revision, then check the work before trusting it. [Back to blog](/blog) Field notes February 22, 2025 · 10 min read Classroom AI · Habits The app list changes every week. The habits do not. Before I care which model a teacher uses, I care whether they can give context, ask for revision, and check the answer before it lands in front of students. The three skills here are the ones I kept coming back to in class: a simple prompting framework, a revision loop, and a way to cite AI use without pretending it did not happen. That is less flashy than a new app list, but it survives the next product launch. Video version: [YouTube](https://youtu.be/NyGZXi49Pdc) Audio version: [Podcast](https://share.transistor.fm/s/8b0ca28f) ### Mindset before skillset Before the skills, the mindset matters. My brain works in analogies, so here is mine for teachers: if you had a student teacher who read every book in the library, every blog online, every textbook about everything, who didn't sleep, who always wanted to make you happy and did things way faster than you could, what jobs would you give them? That assistant would be useful and annoying at the same time. It knows a lot, wants to please you, sounds confident when it is wrong, and rarely asks the question you wish it had asked. How would we use such an assistant? I would start with work I already know how to judge, where I can tell the difference between helpful and wrong. I would not give it a lesson topic I do not understand and let it plan alone. That is the AI habit I want teachers to build: know where your judgment belongs before you hand work to the tool. ### Skill 1: prompting The first habit is giving the tool enough context to be useful. The fastest way I know is a framework simple enough to remember and specific enough to improve the answer. I use a framework called RAFT: - Role: Who would you go to for help on this task? (e.g., a master level history teacher, an expert instructional coach) - Audience: Who are you creating this content for? (e.g., your students, a co-teacher, a mentor) - Format: What format do you want the output in? (e.g., graphic organizer, narrative, bulleted list, table, email) - Topic: What are you talking about? (e.g., the curriculum, the unit, the theme, a certain perspective) Here is an example of a RAFT prompt: Role: Expert instructional coach and curriculum designer Audience: An eighth grade algebra teacher Format: A clear, step-by-step workshop outline Topic: A project-based learning unit on linear equations where students understand real-world applications When you use RAFT, you are doing what you would do with a human helper: giving enough context that the first draft usually lands closer to what you meant. ### Skill 2: iteration No matter how precise your prompt is, don't think the first output is all your gen AI tool can do. Think of it like a first draft from your students. Here are three ways I would revise before I trusted the answer: - "And what else?" Before generating a response, ask the AI tool, "What else do you need to know from me to execute this successfully?" That question catches context you may have missed. It works even better if you have an exemplar, pasted example, or rubric. - Guardrails: Tell the AI what not to do. For example, "Don't use academic words," or "Don't search the internet, just use this source." This further limits the context and improves the output. - Split testing: Ask the AI for multiple different approaches or versions. This gives you a menu of options instead of one "right" answer. ### Skill 3: research and citing Once students can make a decent first draft, citation becomes the next problem. They need to show what source the answer came from and how much help the model gave them. My rule: do not cite the tool itself. Citing "chatgpt .com" is like citing the public library. It is not a specific source. Instead, use an AI tool that can cite the websites it used. Then share the chat log. That gives you the question, the answer, and the handoff in one place. The list is simple: give context, revise the answer, and show your work. The next question is which apps are worth a teacher's time and which ones I would skip. ## Related pages - [All AI notes](/blog) - [Tools](/tools) - [About Aaron](/about) ## Sources and references - [Descript](https://www.descript.com/) - [Perplexity](https://www.perplexity.ai/) --- # AI policy for schools without pretending it is simple Source: https://aaronmakelky.com/blog/ai_policy_for_schools Last updated: 2025-02-17 A school AI policy has to answer normal classroom questions about what students can use and disclose, and where teacher judgment still sits. [Back to blog](/blog) Field notes February 17, 2025 · 8 min read AI Policy · School Administration · Teaching Most school AI policy conversations start too big. People try to solve every classroom, every grade level, every cheating scenario, and every tool update at once. No wonder the document turns into mush. I know because I have had to make this work in a classroom. The policy has to answer normal teacher questions first: what students can use and disclose, and where the teacher still makes the call. This is part two of the AI Teacher Toolkit. I wanted something smaller than a giant district rulebook, something a teacher could use when a student turns in work and the AI question is sitting right there. Video version: [YouTube](https://youtu.be/jGU8jXuIewc) Audio version: [Podcast](https://share.transistor.fm/s/6990f80f) I use four buckets because they keep the conversation from turning into mush: policy, skills, tools, and applications. If one is missing, the classroom plan usually gets weird fast. ### Step one: policy What's my responsibility as a teacher now that AI is here? AI policy works better in layers than as one blanket rule. I think of it like this: State, district, school, then your classroom syllabus. Let's do an example. Does your state currently have AI guidelines? Check your state guidance first, because that number changes too often for an old post to carry the scoreboard by itself. A lot of teachers that I talked to didn't even know their state gave guidelines. Of all the stuff in your state's guidelines, here's the most important thing: How should students be using AI? Here's what Wyoming says: "To prepare all students for the AI rich future, it is imperative that they all learn about AI and have opportunities to learn with AI." AI access has an equity side. Students should understand what the tools can do, where the limits are, and when use is allowed. Why does that matter? If somebody comes at me and says, "How come you're teaching my kid AI?", I say, "Well, have you read my state's guidelines? Because I'm following that guidance." ### The big three I can't write your district policy, but most classroom AI questions still run into the big three: FERPA, CIPA, and COPPA. What do those mean? You have to protect students' identities online. They shouldn't be sharing their name, address, phone number, or email. Your school district has to have a way to filter harmful content from the AI tools. With COPPA, don't have students sign up for accounts if they're not over 13. You can ask AI companies, "Are you guys CIPA and COPPA compliant?" Most teachers I work with say, "I don't have any policy on AI." A lot of them are just leaving it up to the schools and even the individual teachers to figure out. If you're in that boat, ask your administrator the real question: does the lack of policy mean I can't touch the AI topic, or does it mean I can responsibly start doing these things? A flat AI policy breaks fast. Label the assignment instead: what tool use is allowed and disclosed, and where the student still has to do the thinking. ### Transparency: citing AI use The final aspect of AI policy is, how are you going to address transparency? What not to do: Please don't have kids cite the AI tool. That's like citing a library. What I did instead: students turned in links to their chat logs. I started it in 2024, and it changed the use of AI in my classroom. No matter how a student uses AI, they turn in a link to that chat log. It lets you see the questions they asked and where they got stuck, plus how much of it they actually understood. For me, that meant changing some skills and tools without throwing out the fundamentals or the content. That is the policy choice: make AI use visible enough to discuss, or keep pretending the mystery submission is the real problem. I would rather coach the process than grade another submission and guess how much AI was involved. If you want sample syllabus language, the integration scale, or lesson plan ideas, start with [my links page](/links). I keep the classroom resources there because they change faster than a blog post should. ## Related pages - [All AI notes](/blog) - [Tools](/tools) - [About Aaron](/about) ## Sources and references - [Descript](https://www.descript.com/) - [Perplexity](https://www.perplexity.ai/) --- # I made these AI mistakes as a teacher first Source: https://aaronmakelky.com/blog/5-mistakes-teachers-make-with-ai Last updated: 2025-02-02 I made plenty of AI mistakes as a classroom teacher. These are the ones I would clean up first, especially for schools trying to move past panic and tool demos. [Back to blog](/blog) Field notes February 2, 2025 · 15 min read Classroom AI · Teaching This started as the first AI Teacher Toolkit post. It came from mistakes I had already made in the classroom and from conversations with schools, teachers, college professors, and state education teams trying to figure out what generative AI changes in actual classrooms. The five mistakes below are the ones I kept seeing: ignoring AI, trying to ban it, treating AI like an expert you can beat, using it to make more of the same assignments, and getting too loyal to one wrapper. ## Mistake #1: ignoring AI Since November of 2022, students have had ChatGPT on their phone, on their computer. Now there are lots of other ways. But it isn't going away. Ignoring it isn't an option. How do I know? I ask my students. On the second day of class each semester, I give my students an AI use survey. Here's a really interesting finding from over 100 students in the 2024-25 school year: their top used AI tool? Snap AI. Not what most adults think, not what all the techie programmer bros on LinkedIn are talking about. Over 70 percent of my students say they've used Snap AI. That's more than any other tool. Another way we know you can't just ignore AI, is when you ask students, what are you using AI for? School, research, writing, your job? It was none of those things. In fact, 63 percent of my students said they're just using AI for fun. ## Mistake #2: trying to ban or block AI You also can't control what students have on their phones and their personal devices outside of your classroom. Your district or your internet connection might filter chat GPT while they're at school or block certain websites, but you can't control what they put on their phone. The other reason you can't just block or ban AI is because then only the low effort cheaters or the kids that aren't skilled at using it are going to get caught. AI is really an equity issue. The students who know the workarounds will get around the blocks. The students who get caught are usually the ones who needed better coaching, not better surveillance. ## Mistake #3: thinking you are smarter than AI I don't know any teacher who knows more names, dates, and facts than Google does. I don't know any teacher that has a larger repository of information than Wikipedia does. While those tools aren't perfect, they're a lot better at storing information than our brains. And if you try to out Google, Google, or out Wikipedia, Wikipedia, you're not going to win. Don't fight that battle and don't try to be the expert on everything. AI is the same way. If you needed 30 copies of a handout for your students, you could take a pencil and paper and handwrite those out yourself. But you would be crazy to do that. AI is no different. When you're trying to come up with ideas, improve lesson plans, or give students feedback, there is a point where doing it by hand is just wasting time. ## Mistake #4: spamming more of the same Of all these mistakes that I see, number four is probably the most common. I see teachers spamming more of the same. What do I mean by that? What we were teaching a year, two, five years ago, we're just doing the AI version of that. Another thing we get when teachers spam more of the same is what I call AI ping pong. In other words, a student uses an AI tool to write something. That's the ball going over the net. A teacher uses an AI tool to give them feedback. That's hitting the ball back over the net. And it's AI to AI to AI to AI. ## Mistake #5: relying on one wrapper app The fifth mistake I see teachers make with AI, and I have made this mistake myself, is getting too loyal to one wrapper app. In a previous school year, I only used one AI platform with my students. I put all of my content in it. I had a tool to help them write better essays with it. My school did not block it. My kids liked it. It was great. And I thought, look at me. I am so smart. But I was making mistake number five. So what should you do if you do not want to be stuck in one platform? Teach the skill of using AI, not loyalty to one program or one app. I don't teach my kids how to use ChatGPT. They do use it, but I teach habits that transfer there and to the next tool that replaces it. ## Next up: policy The next practical question is policy: not the giant district-document version, but the classroom version of what students can use and disclose, and where the privacy lines matter. If you want the syllabus language, policy samples, lesson ideas, or the student AI survey, start with [my links page](/links). That is where I keep the practical classroom pieces. Anyway, the policy piece is where the weird stuff starts... ## Related pages - [All AI notes](/blog) - [Tools](/tools) - [About Aaron](/about) ## Sources and references - [Descript](https://www.descript.com/) - [Perplexity](https://www.perplexity.ai/)