<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/">
  <channel>
    <title>Aaron Makelky</title>
    <link>https://aaronmakelky.com</link>
    <description>AI field notes, work logs, and tool tests from Aaron Makelky.</description>
    <language>en-us</language>
    <lastBuildDate>Wed, 18 Nov 2026 12:00:00 GMT</lastBuildDate>
    <atom:link href="https://aaronmakelky.com/blog/feed.xml" rel="self" type="application/rss+xml" />
  <item>
    <title>Claude vs Google Workspace: the plan I&apos;d pay for and the apps I&apos;d keep</title>
    <link>https://aaronmakelky.com/blog/claude-vs-google-workspace</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/claude-vs-google-workspace</guid>
    <pubDate>Wed, 18 Nov 2026 12:00:00 GMT</pubDate>
    <description>Ryan Doser and I went through the Claude ecosystem on AI Rabbit Holes: Fable 5.2 vs Opus 5.5, which plan to pay for, Claude&apos;s new whiteboards, docs, and sheets, Claude Duet, and why I wouldn&apos;t cancel Google or Microsoft for any of it.</description>
    <content:encoded>&lt;h2&gt;Which Claude plan Aaron Makelky pays for&lt;/h2&gt;&lt;p&gt;On the November 12, 2026 episode of AI Rabbit Holes, Aaron Makelky said he tells people doing websites, codebases, or other deep work to get the $100 Claude Max plan, which he&amp;apos;s on. Claude doesn&amp;apos;t give a higher multiple at $200, so if you keep hitting limits, he says to try the $200 Ultra plan for a month. For chat, calendar, and meeting notes, he says Meta Muse&amp;apos;s free plan is as good as or better than many $20 plans. Ryan Doser calls Claude Ultra untouchable and says you won&amp;apos;t get enough usage out of Claude&amp;apos;s $20 plan to use all of its features.&lt;/p&gt;&lt;p&gt;On Fable 5.2, Ryan says it isn&amp;apos;t worth it over Opus 5.5 for most tasks, and uses it for high-level work like cleaning up his skills and AGENTS.md. In Aaron&amp;apos;s blind eye doctor test across 13 models, he rated Opus 5.5 higher than Fable 5.2.&lt;/p&gt;&lt;h2&gt;Should you replace Google Workspace with Claude?&lt;/h2&gt;&lt;p&gt;Aaron doesn&amp;apos;t know anybody canceling Microsoft or Google for Claude Docs, Sheets, Slides, or Whiteboards, and says doing it locks you deeper into Claude. Every agent action on a Claude whiteboard uses your rate limits, and Claude doesn&amp;apos;t offer an MCP connector for its own apps, even though Anthropic pioneered MCP. Ryan Doser says the move makes sense for Anthropic as it goes after knowledge work, but expects people to use both.&lt;/p&gt;&lt;p&gt;Aaron&amp;apos;s advice for the Claude ecosystem: use Claude Code projects in the desktop app instead of chat, use Claude Duet for proactive help and scheduled automations, and don&amp;apos;t let your AI tools own your work. Ryan backs up all of his Duet automations to a private GitHub repo every week.&lt;/p&gt;&lt;h2&gt;Questions&lt;/h2&gt;&lt;h3&gt;Which Claude plan does Aaron recommend?&lt;/h3&gt;&lt;p&gt;The $100 Max plan, which is what he&amp;apos;s on. If you&amp;apos;re doing a website, a codebase, or other deep work, you&amp;apos;re using a lot of tokens. Claude doesn&amp;apos;t give you a higher multiple the further up you go, so if you keep hitting limits, he says to go up to the $200 Ultra plan for a month and see. Ryan Doser calls Ultra untouchable, and doesn&amp;apos;t think Claude&amp;apos;s $20 plan is worth it if you plan to use all of its features.&lt;/p&gt;&lt;h3&gt;Is Fable 5.2 worth using over Opus 5.5?&lt;/h3&gt;&lt;p&gt;For most tasks, Ryan Doser says no. He uses Fable 5.2 for high-level work, like cleaning up his skills and his AGENTS.md, and lets Opus, Sonnet, and Haiku do the grunt work. When Aaron let Fable spec out its own subagents, it mostly picked Haiku and Sonnet. In Aaron&amp;apos;s blind test of 13 models, he rated Opus 5.5 higher than Fable 5.2.&lt;/p&gt;&lt;h3&gt;Should you replace Google Docs or Miro with Claude&amp;apos;s artifacts?&lt;/h3&gt;&lt;p&gt;Aaron doesn&amp;apos;t know anybody canceling Microsoft or Google for Claude&amp;apos;s docs, sheets, or whiteboards, and says you&amp;apos;d just get locked deeper into Claude. Every action your agent takes on a board uses your rate limits, Claude doesn&amp;apos;t offer an MCP connector for its own apps, and you can&amp;apos;t take your whiteboard out. He still uses open source whiteboards he can self-host.&lt;/p&gt;&lt;h3&gt;What is the eye doctor test for AI models?&lt;/h3&gt;&lt;p&gt;A blind benchmark Aaron had Fable build. It runs the same task on different models and shows him two outputs at a time, like an eye doctor asking A or B. He picks better, worse, or tied, it reshuffles the winners against other outputs, and it only tells him which model made which at the end. He uses it for things that take taste, like video animations, and says it works for landing pages, UI design, copywriting, or email subject lines.&lt;/p&gt;&lt;h3&gt;Why use Claude Duet instead of Claude Code?&lt;/h3&gt;&lt;p&gt;Ryan Doser&amp;apos;s answer: Duet&amp;apos;s automations run in the cloud on a schedule, which is hard to set up from Claude Code in a terminal. Aaron&amp;apos;s biggest reason is that Duet is proactive. It looks across your connectors and data and tells you when an email looks important or a deliverable isn&amp;apos;t in your Drive yet. He still does his real desktop work in Claude Code projects.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>How to avoid AI slop: start with what a human actually said</title>
    <link>https://aaronmakelky.com/blog/how-to-avoid-ai-slop</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/how-to-avoid-ai-slop</guid>
    <pubDate>Fri, 13 Nov 2026 12:00:00 GMT</pubDate>
    <description>Ryan Doser and I talked about AI slop on AI Rabbit Holes. Why volume stopped working, the anti-slop skills and design rules we use, and why the source of your content decides whether it&apos;s slop.</description>
    <content:encoded>&lt;h2&gt;What Aaron Makelky means by AI slop&lt;/h2&gt;&lt;p&gt;On the November 6, 2026 episode of AI Rabbit Holes, Aaron Makelky defined slop as low effort. Human slop existed before AI, like mass-produced blogging for backlinks, but it was a BB gun. AI is a chain gun that cranks out content at nearly unlimited scale, so winning on volume is easier than ever and worth less. He says to win on value, not volume, and to blame the human using the tool, not the tool.&lt;/p&gt;&lt;p&gt;Ryan Doser defines AI slop as human slop at scale, plus output copied from ChatGPT or Claude with no skills or context behind it. He says AI content is not the same as AI slop.&lt;/p&gt;&lt;h2&gt;How Aaron and Ryan avoid AI slop&lt;/h2&gt;&lt;p&gt;Ryan runs every written output through an anti-slop skill and a brand voice gate, with a second model giving a ship or no ship verdict, plus an anti-spam skill based on Google&amp;apos;s spam policies. Aaron uses his public AI Writing Audit skill, which tags AI writing patterns, corrects them, and keeps a change log, and updates his anti-slop rules whenever he spots a new pattern. He says a good skill says both what not to do and what to do.&lt;/p&gt;&lt;p&gt;For design, Aaron&amp;apos;s agent calls the Impeccable skill automatically through a hook, and a DESIGN.md keeps his video, images, and web consistent. Both start from long-form video transcripts, because content repurposed from something an AI tool wrote stays slop. Aaron&amp;apos;s rule: if you won&amp;apos;t iterate on it from the results, don&amp;apos;t do it.&lt;/p&gt;&lt;h2&gt;Questions&lt;/h2&gt;&lt;h3&gt;Is AI content the same thing as AI slop?&lt;/h3&gt;&lt;p&gt;No. Aaron Makelky compares it to people who wrote with a pencil calling typed work keyboard slop. AI is the medium, and the medium doesn&amp;apos;t make it slop. Slop is low effort. He tells people not to blame the AI tool but the human using it.&lt;/p&gt;&lt;h3&gt;What should an anti-slop skill include?&lt;/h3&gt;&lt;p&gt;Two levers, according to Aaron. One says what not to do, like no delve, and gets every pattern that makes you think &amp;quot;I know this is AI&amp;quot; added the moment you see it. The other nudges the model toward how you do write: your sentence structure, your vocabulary, examples of your writing, or the fact that you like analogies. He says any good skill has some of both, and that it has to keep changing because new models write in new predictable patterns.&lt;/p&gt;&lt;h3&gt;Why does the source of the content matter for AI slop?&lt;/h3&gt;&lt;p&gt;If the original piece came out of an AI tool, Aaron says everything you make from it will be slop: you&amp;apos;re chopping up slop into different shapes. If the source is a transcript of real people talking, like an AI Rabbit Holes episode, the picks and opinions in it carry taste a Google search or Claude wouldn&amp;apos;t pick. Even then, pasting the transcript and asking for a blog with no skills still gives you slop.&lt;/p&gt;&lt;h3&gt;Does Google penalize AI content?&lt;/h3&gt;&lt;p&gt;Ryan Doser says no. Google penalizes spammy, mass-produced AI content, which its spam policies call scaled content abuse. Ryan turned those guidelines into an anti-spam skill that his blog posts go through, alongside his anti-slop skill, and his AI-assisted posts rank on Google.&lt;/p&gt;&lt;h3&gt;Does Aaron use a DESIGN.md for anything besides websites?&lt;/h3&gt;&lt;p&gt;Yes. He uses it for video, images, and web, and says it works for apps too. He isn&amp;apos;t a designer and says he doesn&amp;apos;t understand half of it, but his outputs come out consistent without extra prompting. His agent also has a hook that calls the Impeccable design skill automatically any time it does design work.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>Claude Duet, Meta Muse, dots, or Grok Bot: picking an AI personal assistant</title>
    <link>https://aaronmakelky.com/blog/best-ai-personal-assistant</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/best-ai-personal-assistant</guid>
    <pubDate>Fri, 13 Nov 2026 12:00:00 GMT</pubDate>
    <description>Ryan Doser and I compared the big four AI personal assistants on AI Rabbit Holes. What each one costs, how much usage you really get, where I&apos;d tell someone to start, and why I keep Hermes as insurance.</description>
    <content:encoded>&lt;h2&gt;Which AI personal assistant Aaron Makelky recommends&lt;/h2&gt;&lt;p&gt;On the November 10, 2026 episode of AI Rabbit Holes, Aaron Makelky and Ryan Doser compared Claude Duet, Meta Muse, ChatGPT dots, and Grok Bot. Aaron&amp;apos;s advice: in the subsidized era, use the one you&amp;apos;re already in, and keep a backup. Ryan uses Claude Duet for everything and recommends it to active Claude Code users. Aaron calls Meta Muse the lowest barrier to entry and the one he&amp;apos;d show someone who isn&amp;apos;t technical.&lt;/p&gt;&lt;p&gt;Aaron crossed Grok Bot off the list for now: on the $200 a month plan, its models got worse and its rate limits were cut to about 25 to 30% around November 3, 2026. He finds ChatGPT dots powerful on paper but slow and unreliable. He gets more Meta Muse usage for $16 a month on the web than he gets for $200 a month on Grok Bot.&lt;/p&gt;&lt;h2&gt;Why Aaron keeps Hermes as a backup&lt;/h2&gt;&lt;p&gt;Hermes is the only option besides OpenClaw that lets Aaron change the model and host it anywhere, including a Raspberry Pi. His Grok Bot backs itself up to GitHub every week and Hermes mirrors it, so he could switch to DeepSeek, GLM, OpenAI, or Grok models without starting over. Ryan says most people don&amp;apos;t need Hermes today; both agree it&amp;apos;s insurance for when the subsidies end.&lt;/p&gt;&lt;h2&gt;Questions&lt;/h2&gt;&lt;h3&gt;Which AI personal assistant should a beginner start with?&lt;/h3&gt;&lt;p&gt;Aaron Makelky says to use the one in the ecosystem you&amp;apos;re already in. For someone who isn&amp;apos;t technical, he points to Meta Muse: one chat, a mobile app, a free plan, and a short list of things you can connect it to. Ryan Doser recommends Claude Duet if you already use Claude Code, and Muse as the second option if you&amp;apos;re not in the Anthropic ecosystem.&lt;/p&gt;&lt;h3&gt;Is Grok Bot still worth paying for?&lt;/h3&gt;&lt;p&gt;Not for now, according to Aaron. He&amp;apos;s on the $200 a month plan and says the models got worse and the rate limits were cut to about 25 to 30% of what they were around November 3, 2026. He still loves the app and the voice calling, but he says there&amp;apos;s no way to run your business through it on the highest paid plan.&lt;/p&gt;&lt;h3&gt;Does Claude Duet need a Max plan?&lt;/h3&gt;&lt;p&gt;Duet requires a paid Claude plan. Aaron says Claude Code is about 60 to 70% of his Claude usage and Duet is the other 30 to 40%, and on a Max plan you&amp;apos;re not going to hit rate limits unless you&amp;apos;re already going crazy. Ryan says Duet is very tough to use at scale if you&amp;apos;re not on the $100 or $200 a month plan.&lt;/p&gt;&lt;h3&gt;Why does Aaron still keep Hermes if he uses Duet and Muse?&lt;/h3&gt;&lt;p&gt;It&amp;apos;s his insurance policy. Hermes lets him change the model and host it anywhere, including a Raspberry Pi. His Grok Bot backs itself up to GitHub once a week and Hermes mirrors it, so if he dropped Grok Bot he could pick up where he left off with DeepSeek, GLM, OpenAI, or Grok models. He says to revisit it when the subsidized era ends.&lt;/p&gt;&lt;h3&gt;What are Meta Muse spaces?&lt;/h3&gt;&lt;p&gt;Spaces are small personal apps inside Muse that live in the sidebar and work on mobile and desktop. Aaron compares them to artifacts, but with buttons and actions that can send things through a connector. He uses them for a clip review queue, a fitness and nutrition dashboard that pulls from Apple Health, and a shareable family calendar.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>How to vibe code a real website, and the slide app I built on the same stack</title>
    <link>https://aaronmakelky.com/blog/vibe-code-a-website</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/vibe-code-a-website</guid>
    <pubDate>Sat, 07 Nov 2026 12:00:00 GMT</pubDate>
    <description>Ryan Doser rebuilt a local tire shop&apos;s website live on AI Rabbit Holes from a one-sentence prompt. The stack in plain terms, when to make a skill, why you don&apos;t need an agency or a CMS, and the slide app I built for a talk.</description>
    <content:encoded>&lt;h2&gt;The stack Aaron and Ryan use to vibe code a website&lt;/h2&gt;&lt;p&gt;On the November 5, 2026 episode of AI Rabbit Holes, Ryan Doser redesigned the homepage for Tire Heroes, a mobile tire shop in Des Moines, live with one sentence, four links, and a slash command. The stack: Claude Code or Codex as the brain, Impeccable for design, GitHub to store the code, Astro as the web framework, Cloudflare to deploy, and DataForSEO for keyword research. Aaron Makelky says they arrived at about 95% of the same approach independently.&lt;/p&gt;&lt;p&gt;Aaron&amp;apos;s beginner advice is to build along with the episode, starting with a site for a business you know or your own name.com. He says a skill is worth making for anything you&amp;apos;ll do three times, and that skills belong in GitHub so every AI tool can see them.&lt;/p&gt;&lt;h2&gt;Aaron&amp;apos;s slide app&lt;/h2&gt;&lt;p&gt;For a 21-slide talk at a principals&amp;apos; conference, Aaron built his own presentation app on GitHub and Cloudflare with Claude Code instead of paying 50 to 60 dollars a month for a presenter tool. It has a speaker view, a phone remote with a timer, audience screens that sync and take live votes, and a feedback form saved to a Cloudflare database. The phone remote changes one database row, and the other screens read it.&lt;/p&gt;&lt;p&gt;He says Canva has an amazing remote presentation feature but he hates designing in it, and Figma, where he likes designing, has no remote app.&lt;/p&gt;&lt;h2&gt;Questions&lt;/h2&gt;&lt;h3&gt;What stack do Aaron and Ryan use to vibe code a website?&lt;/h3&gt;&lt;p&gt;Claude Code or Codex as the brain, Impeccable to keep the design from looking like every other AI site, GitHub to store the code, Astro as the web framework, and Cloudflare to deploy it. Ryan adds DataForSEO for keyword research. Aaron says they arrived at about 95% of the same approach without planning it, and that a .com domain for seven to ten dollars is one of the biggest costs.&lt;/p&gt;&lt;h3&gt;Do you need an IDE like VS Code to do this?&lt;/h3&gt;&lt;p&gt;No. Ryan Doser says it&amp;apos;s actually easier in the Claude or Codex desktop app. He prefers VS Code because he can see his skills, folders, and files, keep multiple tabs open, and run Claude Code and Codex in the same window.&lt;/p&gt;&lt;h3&gt;When should something become a skill?&lt;/h3&gt;&lt;p&gt;Aaron&amp;apos;s rule: if you&amp;apos;re going to do it two more times, for a total of three, it should be a skill. Do it once in a normal chat from first prompt to finished product, then tell your coding agent to look back across the chat and turn it into a skill. He saves skills in GitHub so Claude, Codex, or any other tool can see them.&lt;/p&gt;&lt;h3&gt;Does a small business website need a CMS?&lt;/h3&gt;&lt;p&gt;Aaron says if you build websites by dragging boxes, adding widgets, and clicking new page in a sidebar, use a CMS. If you add a page or a blog post with a slash command in Claude, you don&amp;apos;t really need one, because a CMS is an abstraction your AI tool doesn&amp;apos;t need. Ryan doesn&amp;apos;t use one anymore but says it can make sense for bigger teams or an owner who wants to edit things themselves.&lt;/p&gt;&lt;h3&gt;What else can you build on this stack besides a website?&lt;/h3&gt;&lt;p&gt;Aaron built his own slide app for a 21-slide talk at a principals&amp;apos; conference: a speaker view, a phone remote with a timer, and audience screens that sync and take live votes. The phone remote changes one database row, and the other screens read it. It runs on GitHub and Cloudflare, with no AI or API keys needed to run it, instead of a presentation tool that costs 50 to 60 dollars a month.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>The best free GitHub repos for non-coders, and the scan to run first</title>
    <link>https://aaronmakelky.com/blog/best-free-github-repos</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/best-free-github-repos</guid>
    <pubDate>Wed, 04 Nov 2026 12:00:00 GMT</pubDate>
    <description>Ryan Doser and I went through 16 free GitHub repos on AI Rabbit Holes, from SEO and PR to video and writing. Here&apos;s what each one does, which ones I use, and the safety check to run before you fork any of them.</description>
    <content:encoded>&lt;h2&gt;The free GitHub repos Aaron and Ryan recommend&lt;/h2&gt;&lt;p&gt;On the October 29, 2026 episode of AI Rabbit Holes, Ryan Doser picked his free Claude skills, the Google Workspace CLI, Anthropic&amp;apos;s skills repo, last30days, Firecrawl, a Google Search Console MCP, OpenSEO, Newsjack, HyperFrames, and Postiz. Aaron Makelky picked his AI Writing Audit, Herdr, Cloudflare OS, Whisper, Matt Pocock&amp;apos;s skills, and Google&amp;apos;s DESIGN.md, and added Anthropic&amp;apos;s K-12 teacher skills.&lt;/p&gt;&lt;p&gt;Aaron says the single best thing a non-coder can do is make a free GitHub account. He uses stars as bookmarks his AI tool can search, forks repos to customize them, and calls GitHub a backpack for your AI agent that you can take to any other agent or project.&lt;/p&gt;&lt;h2&gt;How to use a public GitHub repo safely&lt;/h2&gt;&lt;p&gt;Ryan copies a public repo&amp;apos;s URL into Claude, ChatGPT, Claude Code, or Codex and says implement this, but only after asking the agent to scan the whole repository for prompt injections, malicious files, or anything that could take API keys or steal data.&lt;/p&gt;&lt;p&gt;Aaron&amp;apos;s steps after forking: scan it for security issues and malicious intent, tell the agent what you run (websites, a blog, landing pages), then ask what to add or take away to fit your use case. He also points out that self-hosting a tool like Postiz isn&amp;apos;t free: you need developer accounts, pay per post, and own the updates and security.&lt;/p&gt;&lt;h2&gt;Questions&lt;/h2&gt;&lt;h3&gt;What should you ask your AI agent to do before you use a public GitHub repo?&lt;/h3&gt;&lt;p&gt;Ryan Doser says to have it scan the entire repository first and look for anything malicious: prompt injections, files that could take your API keys, or anything that could steal your data. Aaron Makelky has seen repos with hidden instructions that older models would follow, like syncing all your data to someone else. Run the scan before you fork it, clone it, or plug it in.&lt;/p&gt;&lt;h3&gt;What&amp;apos;s the difference between starring and forking a GitHub repo?&lt;/h3&gt;&lt;p&gt;Aaron describes a star as a bookmark that doesn&amp;apos;t have to live in your browser or in Notion. You can always find your starred repos, and you can tell your AI tool to look at them. A fork duplicates the repo so it&amp;apos;s your own copy. He forks skill repos, scans them, then asks his agent what to add or take away for his own use case.&lt;/p&gt;&lt;h3&gt;Is self-hosting an open source tool like Postiz free?&lt;/h3&gt;&lt;p&gt;Not really. Aaron points out that posting to a platform like X from a self-hosted scheduler needs a developer account on that platform, and you pay per use, sometimes about a cent a post. You&amp;apos;re also responsible for updates and security issues. He pays for his scheduler because getting his social accounts banned would be a major issue.&lt;/p&gt;&lt;h3&gt;Which free GitHub repos does Aaron use himself?&lt;/h3&gt;&lt;p&gt;His own AI Writing Audit, which checks generated text for common AI writing patterns before anyone reads it; Herdr, for running coding agents across several computers from one terminal; Whisper, for private transcription on his own device; and HyperFrames, which he uses for video editing every day. He also recommended Cloudflare OS, Matt Pocock&amp;apos;s skills, and Google&amp;apos;s DESIGN.md.&lt;/p&gt;&lt;h3&gt;What does Aaron mean by GitHub being a backpack for your AI agent?&lt;/h3&gt;&lt;p&gt;Your skills, instructions, and files live in a GitHub repo instead of inside one AI app, so you can take them to any other AI agent or project. He calls GitHub his gateway into AI coding and building, more than ChatGPT or Claude, and says the single best thing to do is make a free account.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>Cap is the Loom alternative I use, on my own domain</title>
    <link>https://aaronmakelky.com/blog/cap-loom-alternative</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/cap-loom-alternative</guid>
    <pubDate>Sat, 31 Oct 2026 12:00:00 GMT</pubDate>
    <description>Cap has a generous free plan, lets you use your own storage, and puts Loom-style share links on your own domain. Here&apos;s my setup with Cloudflare R2.</description>
    <content:encoded>&lt;h2&gt;Why Aaron uses Cap as his Loom alternative&lt;/h2&gt;&lt;p&gt;Cap is an open source screen recorder for Mac and Windows that records your screen, camera, and mic and gives you a share link, like Loom. It is free for personal use with no time limit on local recordings and share links up to 5 minutes. Cap Pro is $12 a user a month, or $8.16 a month billed yearly ($98), compared with Loom Business at $18 and Business + AI at $24 a user a month.&lt;/p&gt;&lt;p&gt;Aaron Makelky is a paying Cap Pro user, not sponsored. He stores new recordings in his own Cloudflare R2 bucket through Cap Pro&amp;apos;s custom S3 storage and serves share links from caps.aaronmakelky.com.&lt;/p&gt;&lt;h2&gt;How to self-host Cap recordings&lt;/h2&gt;&lt;p&gt;Cap Pro can send new uploads to an S3-compatible bucket such as Cloudflare R2 or AWS S3, or to Google Drive, and can use a custom share domain. Setup: create a private bucket, add Cap&amp;apos;s CORS rule, create a bucket-scoped API token, enter it under Settings, Integrations, S3 Config in Cap, test the connection, then add a custom domain with a CNAME record. Only new uploads move to the bucket.&lt;/p&gt;&lt;p&gt;To self-host everything, Cap&amp;apos;s open source web app runs with Docker Compose, Railway, or Coolify, and Cap Desktop can point at that server.&lt;/p&gt;&lt;h2&gt;Questions&lt;/h2&gt;&lt;h3&gt;Is Cap free?&lt;/h3&gt;&lt;p&gt;Yes, for personal use. The free plan has no time limit on local recordings, includes the Studio Mode editor and 4K export, and gives you shareable links up to 5 minutes long. Commercial use needs a Desktop License ($29 a year) or Cap Pro ($12 a user a month, or $8.16 a month billed yearly).&lt;/p&gt;&lt;h3&gt;Can you self-host Cap?&lt;/h3&gt;&lt;p&gt;Yes, two ways. Cap Pro can send new uploads to your own S3-compatible bucket, like Cloudflare R2 or AWS S3, or to Google Drive, and serve share links from your own domain. Or you can run the whole Cap web app yourself with Docker Compose, Railway, or Coolify. Aaron Makelky uses the first one: Cap Pro, his own Cloudflare R2 bucket, and caps.aaronmakelky.com.&lt;/p&gt;&lt;h3&gt;Is Cap open source?&lt;/h3&gt;&lt;p&gt;Yes. Cap&amp;apos;s code is on GitHub. Most of it is licensed under AGPLv3, and a few camera and screen capture packages use MIT.&lt;/p&gt;&lt;h3&gt;Does the HALLOWEEN30 coupon still work?&lt;/h3&gt;&lt;p&gt;It was good through 11:59 pm on October 31, 2026, for new customers only: Cap Pro for $68.60 a year instead of $98, locked in for as long as the yearly plan stays active. After that, check Cap&amp;apos;s site for a current offer. Students get 30% off.&lt;/p&gt;&lt;h3&gt;How much cheaper is Cap than Loom?&lt;/h3&gt;&lt;p&gt;Loom&amp;apos;s paid plans are $18 (Business) and $24 (Business + AI) a user a month, billed monthly. Cap Pro is $12 a user a month, or $8.16 billed yearly, and it includes AI titles, summaries, chapters, and transcripts.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>Your meeting tool shouldn&apos;t own your transcripts</title>
    <link>https://aaronmakelky.com/blog/your-meeting-tool-shouldnt-own-your-transcripts</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/your-meeting-tool-shouldnt-own-your-transcripts</guid>
    <pubDate>Sat, 31 Oct 2026 12:00:00 GMT</pubDate>
    <description>You aren&apos;t keeping your own copies of meeting notes, but you should. Here&apos;s how: a Granola webhook, a cloud agent, and an Obsidian vault plus Notion database that you own.</description>
    <content:encoded>&lt;h2&gt;How Aaron keeps his own copy of meeting transcripts&lt;/h2&gt;&lt;p&gt;Aaron Makelky keeps every meeting transcript in an Obsidian vault on GitHub and a Notion database he owns, instead of leaving the only copy inside his meeting recorder. Granola sends a webhook to a Grok Bot webhook routine. The same pattern works with a Cursor cloud agent, a Cloudflare Worker, or a Claude or OpenAI agent that exposes HTTPS.&lt;/p&gt;&lt;p&gt;The Granola webhook carries only the event id, event type, note id, and timestamp. The agent verifies the Standard Webhooks signature, answers 2xx within 15 seconds, fetches the full note and transcript from the Granola API, skips notes it already saved with the same content hash, and writes markdown to the vault and a row to Notion. Webhooks need a Granola Business or Enterprise plan.&lt;/p&gt;&lt;h2&gt;Other meeting recorders in the same vault&lt;/h2&gt;&lt;p&gt;A daily Cursor cloud agent at 7pm files Wispr Flow meetings and Google Meet Notes by Gemini into sibling folders with the same markdown shape, and mirrors them into the same Notion database. If a meeting notes app has no webhooks, Aaron&amp;apos;s advice is to set a cloud agent automation that looks for new transcripts one to three times a day.&lt;/p&gt;&lt;h2&gt;Questions&lt;/h2&gt;&lt;h3&gt;Does the Granola webhook include the meeting transcript?&lt;/h3&gt;&lt;p&gt;No. Granola fires note.generated, note.edited, and note.access_granted, and the body only has the event id, event type, note id, and timestamp. Your agent verifies the Standard Webhooks signature, answers 2xx within 15 seconds, then fetches the full note from GET https://public-api.granola.ai/v1/notes/{id}?include=transcript with your Granola API key.&lt;/p&gt;&lt;h3&gt;Which Granola plans support webhooks?&lt;/h3&gt;&lt;p&gt;Business or Enterprise. Personal plans cannot create a webhook endpoint.&lt;/p&gt;&lt;h3&gt;What if my meeting notes app does not have webhooks?&lt;/h3&gt;&lt;p&gt;Set an automation in your cloud agent of choice to look for new transcripts one to three times a day. Aaron runs a daily Cursor cloud agent at 7pm that files Wispr Flow and Google Meet Notes by Gemini into the same Obsidian vault and Notion database.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>Build it or buy it: the SaaS tools I&apos;d replace with AI, and the ones I still pay for</title>
    <link>https://aaronmakelky.com/blog/build-or-buy-saas</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/build-or-buy-saas</guid>
    <pubDate>Fri, 30 Oct 2026 12:00:00 GMT</pubDate>
    <description>Ryan Doser and I went category by category on AI Rabbit Holes: website builders, CRMs, email, schedulers, Zapier, design, video editing, notes, and meeting recorders. Here&apos;s which ones I&apos;d vibe code and which ones I&apos;d keep paying for.</description>
    <content:encoded>&lt;h2&gt;How Aaron decides whether to build or buy a SaaS tool&lt;/h2&gt;&lt;p&gt;On the October 27, 2026 episode of AI Rabbit Holes, Ryan Doser asked three questions before replacing a SaaS tool with AI: can my AI agent already do this, is it a whole job or just a button, and when did I last use it. Aaron Makelky added a fourth: who is it for? A tool only he and his agents use is easier to build, because he doesn&amp;apos;t care about the buttons, the design, or the human user experience.&lt;/p&gt;&lt;p&gt;Aaron&amp;apos;s closing rule: if it&amp;apos;s not live and collaborative, you can probably build it yourself, and then you decide whether it&amp;apos;s worth the time and the tokens. He also checks how AI friendly a tool is. If it isn&amp;apos;t adding an MCP or letting you connect Claude to it, he starts thinking about a replacement.&lt;/p&gt;&lt;h2&gt;Aaron&amp;apos;s build or buy calls by category&lt;/h2&gt;&lt;p&gt;Build or skip: AI website builders (move the site to GitHub and Cloudflare and pay seven to ten dollars a year for the domain), Kajabi-style standalone course sites (he had Cursor clone one in under an hour), Zapier and n8n (run automations on the cloud agent you already pay for, like ChatGPT Work, Claude, or Grok Bot), and AI writing tools (the same models with a system prompt).&lt;/p&gt;&lt;p&gt;Keep paying or use a free plan: Cal.com&amp;apos;s free plan for scheduling, a free Notion database as a CRM, Beehiiv or Kit for email deliverability, Buffer for its MCP, Krea and HeyGen for images, video, and voice, Figma for live whiteboards, CapCut Pro on his phone, Riverside for live podcast recording, free Notion and Obsidian for notes, Linear as a context layer for his agents, and Wispr Flow at $72 a year for meeting notes and dictation on his iPhone and Mac.&lt;/p&gt;&lt;h2&gt;Questions&lt;/h2&gt;&lt;h3&gt;What should you ask before vibe coding your own version of a SaaS tool?&lt;/h3&gt;&lt;p&gt;Ryan Doser asks three questions: can my AI agent already do this, is it a whole job or just a button, and when did I last use this tool. Aaron Makelky adds a fourth: who is it for? A tool only you and your agents use is much easier to build, because you don&amp;apos;t care about the buttons, the design, or the human user experience.&lt;/p&gt;&lt;h3&gt;Which SaaS tools does Aaron still pay for?&lt;/h3&gt;&lt;p&gt;Email marketing, because you&amp;apos;re paying a company like Beehiiv or Kit to handle deliverability. Live, collaborative podcast recording, like Riverside. Figma, for live workshop whiteboards. CapCut Pro, for editing on his phone. And Wispr Flow, at $72 a year, because he wants the same dictation and meeting notes on his iPhone and his Mac.&lt;/p&gt;&lt;h3&gt;Can you use Notion as a free CRM?&lt;/h3&gt;&lt;p&gt;Aaron says yes for most small businesses. A Notion database works like a giant spreadsheet, and with Claude Code or Codex connected you can ask for contact info, a rating on where each lead came in, categories, and a most-recent-first sort. He wouldn&amp;apos;t build a CRM web app, and he doesn&amp;apos;t expect a company already on HubSpot or Salesforce to switch.&lt;/p&gt;&lt;h3&gt;Do you still need Zapier or n8n?&lt;/h3&gt;&lt;p&gt;Aaron hasn&amp;apos;t touched one in years. When people learned on those tools, not every app had an API and there weren&amp;apos;t cloud agents. Now most people already pay for a tool like ChatGPT Work, Claude, or Grok Bot that has a virtual machine, which can run automations in the cloud without a laptop open. Ryan Doser says Zapier and n8n still have a moat with enterprise companies.&lt;/p&gt;&lt;h3&gt;When should you replace a SaaS tool?&lt;/h3&gt;&lt;p&gt;Aaron looks at how AI friendly the tool is making its product. If it isn&amp;apos;t adding an MCP or letting you connect Claude to it, start thinking about a different tool or your own build. His other rule: if it&amp;apos;s not live and collaborative, you can probably build it yourself, and then you decide whether that&amp;apos;s worth the time and the tokens.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>The Lovable tax, and the $10 a year website stack I use instead</title>
    <link>https://aaronmakelky.com/blog/lovable-alternative</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/lovable-alternative</guid>
    <pubDate>Fri, 30 Oct 2026 12:00:00 GMT</pubDate>
    <description>Ryan Doser and I talked about his viral Lovable post on AI Rabbit Holes. Why I left Lovable for GitHub and Cloudflare, what a site costs on our stack, and my chase it or ignore it calls on this week&apos;s AI hype.</description>
    <content:encoded>&lt;h2&gt;Why Aaron left Lovable&lt;/h2&gt;&lt;p&gt;Aaron Makelky used Lovable in the summer of 2025. A friend asked where his back end and his code were, and told him to get off Lovable once he knew he wanted to make websites. He moved his code to a GitHub repo, deployed his site on Cloudflare, and canceled his Lovable account. He says people can argue the word scam, but you pay a tax using a tool like Lovable.&lt;/p&gt;&lt;p&gt;Aaron and Ryan Doser each arrived on their own at the same stack: an AI coding agent like Claude Code or Codex, GitHub for the code, the Astro web framework, and Cloudflare to deploy. By Aaron&amp;apos;s math a site costs about $10 a year, mostly the domain, against $25 a month for Lovable Pro. Because the code lives in GitHub, switching from Claude Code to Codex changes nothing for his sites.&lt;/p&gt;&lt;h2&gt;Aaron&amp;apos;s chase it or ignore it calls&lt;/h2&gt;&lt;p&gt;On the October 15, 2026 episode of AI Rabbit Holes, Aaron said to chase Wabi for multiplayer group chats with your agent, the X Subscription for people already paying for Grok, Grok Bot, Cursor, or X, and Claude Code projects for complex work with subagents.&lt;/p&gt;&lt;p&gt;He said to ignore Gemini 4 Argon, which falls short of Claude Opus 5.5 and is moving pro models to a higher plan; Muse Wonder, whose full model needs a separate Muse Code plan; Instinct, over privacy; and Hark, whose credits list no model or token count.&lt;/p&gt;&lt;h2&gt;Questions&lt;/h2&gt;&lt;h3&gt;Is Lovable a scam?&lt;/h3&gt;&lt;p&gt;Aaron Makelky says people can argue the word scam, but you at least pay a tax using a tool like Lovable versus building the site yourself with Claude Code or Codex, GitHub, Astro, and Cloudflare. He used Lovable in the summer of 2025, then moved his code to a GitHub repo, deployed on Cloudflare, and canceled his account.&lt;/p&gt;&lt;h3&gt;What does Aaron use instead of Lovable?&lt;/h3&gt;&lt;p&gt;The same stack Ryan Doser uses, which they each arrived at on their own: an AI coding agent like Claude Code or Codex, GitHub for the code, the Astro web framework, and Cloudflare to deploy. Because the code lives in his GitHub, he can switch from Claude Code to Codex and nothing changes for his websites.&lt;/p&gt;&lt;h3&gt;How much does a website cost on that stack?&lt;/h3&gt;&lt;p&gt;About $10 a year, by Aaron&amp;apos;s math. The AI subscription is one you already have, most .com domains are seven to ten dollars a year, GitHub and Astro are free, and you deploy on a free Cloudflare plan. Ten leads at $10 each is a ten to one return, and he says you could make that a thousand to one.&lt;/p&gt;&lt;h3&gt;Can a site built with AI rank on Google?&lt;/h3&gt;&lt;p&gt;Ryan Doser&amp;apos;s Des Moines Septic site was built 100% with AI on this stack, and on the episode it ranked number one for septic tank pumping in Des Moines, ahead of the real septic companies, with three to five leads a week. Ryan&amp;apos;s own site has driven over $50,000 in digital product sales since he rebuilt it this way.&lt;/p&gt;&lt;h3&gt;Which AI tools did Aaron say to chase this week?&lt;/h3&gt;&lt;p&gt;Wabi, because it&amp;apos;s multiplayer and you can bring your agent into a group chat; the X Subscription, for most people already paying for Grok, Grok Bot, Cursor, or X; and Claude Code projects, if you do complex work with subagents. He said to ignore Gemini 4 Argon, Muse Wonder, Instinct, and Hark.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>My video review notes live in GitHub now</title>
    <link>https://aaronmakelky.com/blog/video-review-desk</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/video-review-desk</guid>
    <pubDate>Tue, 27 Oct 2026 12:00:00 GMT</pubDate>
    <description>I built a review page for AI Rabbit Holes clips where every note is a JSON file pinned to a clip, a version, and a second, committed to GitHub. Here are my two real notes on clip 2, what they fixed, and the open-source tool.</description>
    <content:encoded>&lt;h2&gt;Video review notes as files&lt;/h2&gt;&lt;p&gt;Aaron Makelky reviews AI Rabbit Holes clips on a page where each note is a JSON file pinned to a clip, a version, and a second, committed to git under his name. Anything that can read the repo, from a teammate on github.com to Codex, can list open notes and resolve them with the version that fixed them.&lt;/p&gt;&lt;p&gt;The tool is open source under the MIT license at github.com/a-makelky/video-review-desk: one Python file that serves the review page and the videos on localhost.&lt;/p&gt;&lt;h2&gt;Questions&lt;/h2&gt;&lt;h3&gt;What is video-review-desk?&lt;/h3&gt;&lt;p&gt;An open-source review page for video versions. You pause a clip, type a note, and it is saved as a JSON file pinned to that clip, version, and second. It is one Python file, MIT licensed, at github.com/a-makelky/video-review-desk.&lt;/p&gt;&lt;h3&gt;Why keep video review notes in a GitHub repo?&lt;/h3&gt;&lt;p&gt;Anything that can read the repo can read and answer the notes: a teammate on github.com, a script, or a coding agent like Codex or Claude Code. Each note is its own file and its own commit, so two people can leave notes at once without a merge conflict.&lt;/p&gt;&lt;h3&gt;How do you cut podcast clips from a Riverside export?&lt;/h3&gt;&lt;p&gt;The Riverside timeline export is a zip with each camera file, each mic WAV, the screen share, and an XML timeline. Every camera in the XML starts at the session clock, so transcript word timings line up with both cameras. The clip pipeline reads only the seconds each clip needs straight out of the zip.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>How to automate the non-creative parts of your creative job</title>
    <link>https://aaronmakelky.com/blog/automate-non-creative-work</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/automate-non-creative-work</guid>
    <pubDate>Tue, 27 Oct 2026 12:00:00 GMT</pubDate>
    <description>It can probably do most of your repetitive, tedious, logistical, garbage work, and you don&apos;t need technical chops to set it up.</description>
    <content:encoded>&lt;h2&gt;What to hand to AI&lt;/h2&gt;&lt;p&gt;Aaron Makelky automates the repetitive, manual, and mechanical parts of a creative job. A chatbot and about an hour of connecting the apps he already uses is enough. He still writes the copy and builds the slides himself.&lt;/p&gt;&lt;p&gt;Uploading a video used to take about 30 minutes, 7-10 times a week: copy the Notion doc onto a social calendar, paste the copy into the scheduler, then mark it published. Typing &amp;quot;/calendar&amp;quot; in Notion now has Notion AI create that calendar entry, except the copy, in a few seconds. He wrote the instruction in Notion&amp;apos;s AI. Claude or ChatGPT can take the same kind of instruction. Next he wants Notion AI connected to Sprout Social, once Sprout&amp;apos;s API can handle it.&lt;/p&gt;&lt;h2&gt;Linear, files, and where to start&lt;/h2&gt;&lt;p&gt;Linear is the project tool he does not want to open. A Claude automation checks it the hour before he logs in, finds tasks that are overdue or due in the next two weeks, prioritizes them, and schedules time with the project links in the calendar entry. Typing &amp;quot;document the work&amp;quot; adds the links and marks the task done. He built it in Claude Cowork. Codex can do the same job in ChatGPT. Cowork also pulls follow-ups from a Notion meeting transcript into Linear and reads his Gmail each morning.&lt;/p&gt;&lt;p&gt;His decks live in Google Slides, Canva, and Figma, including piles of final_v4 copies. He describes the deck and the change to Claude. Claude finds the deck and the slides and returns a punch list in under a minute. He still makes the slides himself. To start, pick one non-creative thing you don&amp;apos;t like doing, or that you struggle to find time for, ask a chatbot how to automate it, and follow its directions.&lt;/p&gt;&lt;h2&gt;Questions&lt;/h2&gt;&lt;h3&gt;What parts of a creative job should you automate with AI?&lt;/h3&gt;&lt;p&gt;The repetitive, manual, or mechanical parts. Aaron&amp;apos;s examples are the data entry of uploading a video, sorting tasks in a project tool, and hunting down an old slide. He still writes the copy and builds the slides himself.&lt;/p&gt;&lt;h3&gt;Do you need to code to set these automations up?&lt;/h3&gt;&lt;p&gt;No. A chatbot and about an hour of connecting the apps you already use is enough. Aaron wrote one instruction in Notion AI. Claude Cowork handles the Linear and file jobs, and Codex can do the same work in ChatGPT.&lt;/p&gt;&lt;h3&gt;Where should you start?&lt;/h3&gt;&lt;p&gt;Pick one non-creative thing you don&amp;apos;t like doing, or that you struggle to find time for. Ask your chatbot how to automate it, and follow its directions.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>Your AI app should not own your work</title>
    <link>https://aaronmakelky.com/blog/your-ai-app-should-not-own-your-work</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/your-ai-app-should-not-own-your-work</guid>
    <pubDate>Mon, 26 Oct 2026 12:00:00 GMT</pubDate>
    <description>When I canceled ChatGPT Plus, I realized I&apos;d left my entire working life in it. The binder method keeps my instructions, context, and work in one GitHub folder any AI app can use, so I can switch apps any time. Copy the prompt to set up yours.</description>
    <content:encoded>&lt;h2&gt;The binder method&lt;/h2&gt;&lt;p&gt;Aaron Makelky keeps all of his AI work in a binder: a GitHub repo of plain markdown files that ChatGPT, Claude, Cursor, Codex, and open-source models can all check out, read, and update. GitHub records what each app changed, so he can see which one made a mess.&lt;/p&gt;&lt;p&gt;His binder has five tabs: a running to-do list called today, a projects folder, an AI-agent instructions folder, a skills folder with reusable instructions and app memory, and a checkout card that tracks which agent on which device touched his work. A prompt template in his gist walks any AI app through building one.&lt;/p&gt;&lt;h2&gt;Questions&lt;/h2&gt;&lt;h3&gt;What is the binder method?&lt;/h3&gt;&lt;p&gt;Keeping all your AI instructions, context, documentation, and work records in plain markdown files in one folder, usually a GitHub repo, that any AI app can read and update. The app becomes a tool for getting work done, not the place where the work lives.&lt;/p&gt;&lt;h3&gt;Why use GitHub for the binder?&lt;/h3&gt;&lt;p&gt;GitHub records what changed and when. Each AI app checks the files out and back in, so if one breaks an automation or posts something you did not ask for, you can see which one it was and what went wrong. It is free for text files.&lt;/p&gt;&lt;h3&gt;How do I set up my own binder?&lt;/h3&gt;&lt;p&gt;Paste the prompt from the gist into ChatGPT, Claude, Cursor, Codex, or another AI app. It interviews you about your work, then builds the folder: a today.md to-do list, project notes, agent instructions, skills, and a handoff note.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>How to use Cursor Automations, with a prompt to design your own</title>
    <link>https://aaronmakelky.com/blog/how-to-use-cursor-automations</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/how-to-use-cursor-automations</guid>
    <pubDate>Fri, 16 Oct 2026 12:00:00 GMT</pubDate>
    <description>Cursor Automations are cloud agents that run on a schedule with your computer off. How Aaron uses them to get more out of Cursor, and one prompt to paste in and design your own.</description>
    <content:encoded>&lt;h2&gt;How to use Cursor Automations&lt;/h2&gt;&lt;p&gt;Aaron Makelky uses Cursor Automations, Cursor&amp;apos;s cloud agents, to file his inbox, meeting notes, and new business into a repo and to write a morning brief before he starts work. A Cursor Automation runs on a schedule or when something happens, such as a new email or a merged pull request, and it does not need his computer to be on.&lt;/p&gt;&lt;p&gt;His post gives one prompt to paste into Cursor. It covers five shapes: inbox triage, a chief-of-staff repo, meeting notes, lead intake and enrichment, and a morning briefing. Before it writes anything, it asks which tools you use and where results should land.&lt;/p&gt;&lt;h2&gt;Questions&lt;/h2&gt;&lt;h3&gt;How do I use Cursor Automations?&lt;/h3&gt;&lt;p&gt;Open cursor.com/automations, then create one automation for one repeated job. Give it a trigger, either a schedule or an event such as a new email or a merged pull request, and tell it where the result should go. Or paste the prompt from this post and let Cursor design three for you.&lt;/p&gt;&lt;h3&gt;Are Cursor Automations the same as cloud agents?&lt;/h3&gt;&lt;p&gt;Yes. Cursor calls them automations, and the same feature is what Aaron Makelky describes as cloud agents: tasks that trigger on a schedule and run in the cloud without your device needing to be on.&lt;/p&gt;&lt;h3&gt;What is a Cursor Automation?&lt;/h3&gt;&lt;p&gt;A cloud agent in Cursor that runs on its own, on a schedule or when something happens, such as a new email, a webhook, or a merged pull request. It runs in the cloud, so your own computer does not need to be on.&lt;/p&gt;&lt;h3&gt;Why does the prompt ask questions before it designs anything?&lt;/h3&gt;&lt;p&gt;It asks which tools you already use, which check you repeat by hand, where results should land, and what must stay in your hands. Your answers decide the automations, so the prompt does not invent your stack.&lt;/p&gt;&lt;h3&gt;What stays with me when an automation runs?&lt;/h3&gt;&lt;p&gt;Sending, posting, spending, booking, and deleting. The prompt tells the agent to leave those actions to you, except a morning brief you asked to receive by email or Slack.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>Your AI tools will use GitHub more than you do</title>
    <link>https://aaronmakelky.com/blog/github-for-marketers</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/github-for-marketers</guid>
    <pubDate>Fri, 16 Oct 2026 12:00:00 GMT</pubDate>
    <description>Ryan Doser and I talked GitHub for marketers on AI Rabbit Holes. Why I keep my work in GitHub instead of Google Drive, how I back up my Grok Bot setup, and the chief of staff repo I check out like a binder.</description>
    <content:encoded>&lt;h2&gt;How Aaron uses GitHub as a non-developer&lt;/h2&gt;&lt;p&gt;Aaron Makelky started using GitHub seriously over Christmas break at the end of 2025, and his GitHub contribution graph shows 3,610 contributions in the last year. He thinks of a repo as a backpack or briefcase of important files that ChatGPT, Claude, and Grok Bot can all use, and says his AI tools use GitHub more than he does.&lt;/p&gt;&lt;p&gt;He keeps work in GitHub so canceling a tool like Lovable doesn&amp;apos;t take his website with it. A bot in Grok Bot maps all of his bots and automations into a GitHub repo once a week, so he can fix a broken setup or recreate it in Hermes. He prefers GitHub to Google Drive because his AI tools are bad at using Drive&amp;apos;s connectors.&lt;/p&gt;&lt;h2&gt;Forking repos and the chief of staff repo&lt;/h2&gt;&lt;p&gt;Aaron forks public repos and tweaks them, or has his AI tool build its own version from the same principles. From a skills repo with 200 skills he might want three. Ryan Doser warns that some public repos carry prompt injections that can steal API keys.&lt;/p&gt;&lt;p&gt;Aaron&amp;apos;s chief of staff repo holds his to-do list, projects, priorities, and skills. He checked it out from GitHub every morning like a binder and pushed it back every night, and his AI binder method page has a prompt to set it up. Ryan Doser runs seven websites from GitHub with Claude Code, Codex, Astro, and Cloudflare.&lt;/p&gt;&lt;h2&gt;Questions&lt;/h2&gt;&lt;h3&gt;Do marketers really need to learn GitHub?&lt;/h3&gt;&lt;p&gt;Aaron says it&amp;apos;s going to become like water for non-developers, so you&amp;apos;ll end up using it eventually, and you don&amp;apos;t need to be deep in the weeds. Ryan Doser says you only need three words: repository, commit, and push. For anything else, Aaron asks his AI tool what it just did, where it put something, or what a word means.&lt;/p&gt;&lt;h3&gt;Why use GitHub instead of Google Drive or Dropbox?&lt;/h3&gt;&lt;p&gt;Aaron&amp;apos;s AI tools aren&amp;apos;t very good at using Google Drive, and its connectors are limited. He couldn&amp;apos;t even get them to download videos from it. GitHub is built for AI tools to put files in, pull them out, and sync them. Ryan used to back up to Google Drive, but each backup took three or four hours and the storage kept costing more.&lt;/p&gt;&lt;h3&gt;How does Aaron back up his Grok Bot setup?&lt;/h3&gt;&lt;p&gt;One of his bots maps every bot and every automation he has in Grok Bot into a GitHub repo, once a week, on an automation he never touches. If he breaks something, it&amp;apos;s easy to fix. If he cancels Grok Bot, he can point Hermes or another assistant at the repo and say this is how I used to use Grok Bot, set yourself up just like this.&lt;/p&gt;&lt;h3&gt;What is a chief of staff repo?&lt;/h3&gt;&lt;p&gt;A GitHub repo with your to-do list, your projects, what you&amp;apos;re working on, and some skills. Aaron checked it out every morning like a binder, worked in it with his AI tool, and pushed it back every night. Because he owns it, he can plug it into Claude, ChatGPT, Grok Bot, or Meta&amp;apos;s assistant. His AI binder method page walks through the setup with a prompt.&lt;/p&gt;&lt;h3&gt;Should you use public GitHub repos as they are?&lt;/h3&gt;&lt;p&gt;Aaron rarely does. He forks a repo, sees how it works, and tweaks it, or gives the link to his AI tool and has it build his own version from the same principles. A skills repo might have 200 skills when he wants three. Ryan Doser warns that some public repos carry prompt injections that can steal your API keys.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>I gave every AI agent its own Linear account</title>
    <link>https://aaronmakelky.com/blog/linear-multi-agent-workspace</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/linear-multi-agent-workspace</guid>
    <pubDate>Tue, 13 Oct 2026 12:00:00 GMT</pubDate>
    <description>Cursor, Codex, Hermes, Grok Bot, and Devin all take work from one Linear workspace. How I set it up with one rules doc, one OAuth app per agent, and a webhook gateway, plus a prompt your agent can use to walk you through yours.</description>
    <content:encoded>&lt;h2&gt;How Aaron&amp;apos;s Linear workspace is organized&lt;/h2&gt;&lt;p&gt;Aaron Makelky runs his work and his AI agents out of one Linear workspace. Every issue belongs to a project and every project belongs to one of five initiatives: Consulting, Content, Property, Personal, and Systems &amp;amp; Agents.&lt;/p&gt;&lt;p&gt;One Linear document, his agent rules, tells every agent how to work: search before creating an issue, one outcome per issue, a verb-first title, a priority, Aaron as assignee and the agent as delegate, no due dates, In Progress when starting, In Review when finished, and no email, posting, spending, or contacting people without Aaron&amp;apos;s approval.&lt;/p&gt;&lt;h2&gt;OAuth apps and webhooks&lt;/h2&gt;&lt;p&gt;Cursor, Codex cloud, and Devin use their native Linear integrations. Aaron made one Linear OAuth application each for Codex Local and Hermes Local, which run on his own computers, and for Grok Bot. Each is installed with actor=app and the app:assignable and app:mentionable scopes, so each agent acts in Linear under its own name.&lt;/p&gt;&lt;p&gt;Each app sends Agent session events to a Cloudflare Worker gateway. The gateway checks the HMAC-SHA256 Linear-Signature with that app&amp;apos;s signing secret, queues the event, posts a Received receipt, and lets a runner on Aaron&amp;apos;s own computer claim it. The runner&amp;apos;s answer goes back into the same Linear session and the issue moves to In Review.&lt;/p&gt;&lt;h2&gt;Questions&lt;/h2&gt;&lt;h3&gt;Do AI agents cost a seat in Linear?&lt;/h3&gt;&lt;p&gt;No. Agents installed in a Linear workspace as apps don&amp;apos;t count as billable users. Aaron has six agent apps in his workspace and pays for one seat, his own.&lt;/p&gt;&lt;h3&gt;What is actor=app in a Linear OAuth app?&lt;/h3&gt;&lt;p&gt;It is a parameter on the OAuth authorization URL. With actor=app, Linear installs the application as its own member of the workspace, with its own name and icon, instead of acting as the person who installed it. That is what makes an agent show up as Codex Local or Hermes Local instead of as you.&lt;/p&gt;&lt;h3&gt;Which scopes does a Linear agent app need?&lt;/h3&gt;&lt;p&gt;app:assignable so you can delegate issues to it, and app:mentionable so you can @mention it. Add read, and write if the agent posts replies or changes issues. An app installed with actor=app cannot also request the admin scope.&lt;/p&gt;&lt;h3&gt;How does a Linear webhook start my agent?&lt;/h3&gt;&lt;p&gt;When you delegate an issue to the agent or @mention it, Linear starts an agent session and sends a signed POST to the app&amp;apos;s webhook URL. Your receiver checks the HMAC-SHA256 signature, answers within 5 seconds, posts a first thought within 10 seconds, and hands the task to wherever the agent runs.&lt;/p&gt;&lt;h3&gt;Do I need webhooks for Cursor, Codex, or Devin?&lt;/h3&gt;&lt;p&gt;No. Cursor, Codex cloud, and Devin have native Linear integrations you install from Linear. You only build an OAuth app and a webhook receiver for an agent that runs somewhere you control, like a CLI on your own computer.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>Opus 5.5, Luna, and Muse: picking an AI daily driver</title>
    <link>https://aaronmakelky.com/blog/claude-opus-5-5-best-ai-model</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/claude-opus-5-5-best-ai-model</guid>
    <pubDate>Sat, 10 Oct 2026 12:00:00 GMT</pubDate>
    <description>Ryan Doser tested Claude Opus 5.5, GPT-6 Sol, and Muse Spark on a landing page and a launch email. I still use Luna as my daily driver, and Muse surprised me.</description>
    <content:encoded>&lt;h2&gt;Aaron&amp;apos;s take on the new models&lt;/h2&gt;&lt;p&gt;Aaron Makelky thinks the real fight between Anthropic and OpenAI is the most useful daily driver model, not the top benchmark. Fable and GPT-6 Astra use up rate limits and cost a lot on the API, so most work needs a faster, cheaper model.&lt;/p&gt;&lt;p&gt;His daily driver is GPT-6 Luna, which he runs for background and parallel tasks while a stronger orchestrator model plans. Ryan Doser&amp;apos;s daily driver is Claude Opus 5.5, which he says raised quality over Opus 5 and uses tokens more efficiently.&lt;/p&gt;&lt;p&gt;On Ryan&amp;apos;s landing page test, Aaron preferred some details in Muse Spark&amp;apos;s page over Opus 5.5 and said Opus 5.5 beat Fable 5.1&amp;apos;s page from the week before. On the launch email, Ryan picked Opus 5.5, with GPT-6 Sol a close second.&lt;/p&gt;&lt;h2&gt;Muse, the ChatGPT desktop app, and skills&lt;/h2&gt;&lt;p&gt;Aaron uses Meta&amp;apos;s Muse every day on its free tier. It turned a photo of a birthday invite into a calendar event and called a doctor&amp;apos;s office for him, then patched him through to the receptionist. You can&amp;apos;t choose or see the model it runs on.&lt;/p&gt;&lt;p&gt;He uses the ChatGPT desktop app for its in-app browser with cookies and extensions and for computer use, and Herdr to run terminal sessions across machines.&lt;/p&gt;&lt;p&gt;With custom GPTs retiring on December 11, 2026, Aaron and Ryan both use portable skills. Aaron&amp;apos;s kids sports upload skill uploads, labels, and publishes his kids&amp;apos; game film as unlisted YouTube videos.&lt;/p&gt;&lt;h2&gt;Questions&lt;/h2&gt;&lt;h3&gt;What does Aaron mean by a daily driver AI model?&lt;/h3&gt;&lt;p&gt;The model he uses for most day-to-day work. Top models like Fable and GPT-6 Astra use up rate limits fast and get expensive on the API, so for most tasks Aaron wants a faster, cheaper model that&amp;apos;s still good enough.&lt;/p&gt;&lt;h3&gt;Which model does Aaron use as his daily driver?&lt;/h3&gt;&lt;p&gt;GPT-6 Luna. He ran 5.6 Luna on the max or ultra setting more than any other model, and he uses Luna 6 for background and parallel jobs he doesn&amp;apos;t need fast, while a stronger model writes the plan. Ryan Doser&amp;apos;s daily driver is Claude Opus 5.5.&lt;/p&gt;&lt;h3&gt;Was Opus 5.5 better than Muse Spark on Ryan&amp;apos;s landing page test?&lt;/h3&gt;&lt;p&gt;Not on everything. Aaron preferred Muse&amp;apos;s glow effect on the checkout button and its single row of full-color client logos, but wouldn&amp;apos;t call it better overall. He said Opus 5.5 beat what Fable 5.1 did the week before. On the launch email, Ryan picked Opus, with GPT-6 Sol a close second.&lt;/p&gt;&lt;h3&gt;What is Meta&amp;apos;s Muse good for?&lt;/h3&gt;&lt;p&gt;Everyday digital tasks for people who don&amp;apos;t want to touch GitHub or Cursor. Aaron used it to turn a photo of a birthday invite into a calendar event and gift reminder, to analyze his Instagram reels, and to call a doctor&amp;apos;s office, wait through the phone tree, and patch him through to the receptionist.&lt;/p&gt;&lt;h3&gt;What should I do about custom GPTs going away?&lt;/h3&gt;&lt;p&gt;Aaron replaced a custom GPT use case with a skill, a markdown file with steps, which uploads and labels his kids&amp;apos; game film. If you try OpenAI&amp;apos;s plugins instead, he says to keep the work in a markdown file or a GitHub repo you can take with you.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>I use open source AI models for the work I hate</title>
    <link>https://aaronmakelky.com/blog/open-source-ai-models</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/open-source-ai-models</guid>
    <pubDate>Sat, 10 Oct 2026 12:00:00 GMT</pubDate>
    <description>Ryan Doser and I talked open source AI models on AI Rabbit Holes. A 3 GB Qwen model renames my screenshots offline, GLM Flash runs my personal agent in Hermes, and OpenRouter keeps it all on one API key.</description>
    <content:encoded>&lt;h2&gt;Where Aaron uses open source AI models&lt;/h2&gt;&lt;p&gt;Aaron Makelky runs a 3 GB Qwen image model on his Mac to rename his daily screenshots offline, and Gemma models in Google AI Edge Gallery on his iPhone for questions about medical records, banking, and personal topics that he doesn&amp;apos;t want leaving his device.&lt;/p&gt;&lt;p&gt;In the cloud, he uses the Hermes desktop app from Nous Research as the interface for open source models. His default is GLM 5.3 Flash on a Z.ai coding plan, and the Hermes model picker also holds his ChatGPT models, MiniMax, and the Nous Portal list, with side-by-side answers from several models on one prompt.&lt;/p&gt;&lt;p&gt;He uses OpenRouter so one API key covers every provider, and to try a new model on a small project for a dollar or fifty cents.&lt;/p&gt;&lt;h2&gt;What he recommends&lt;/h2&gt;&lt;p&gt;Give open source models the day-to-day background tasks: searches, finding files, drafts, calendars, and logging. Aaron has defaulted to GLM Flash models for at least nine months and points to DeepSeek for price.&lt;/p&gt;&lt;p&gt;Hosting the biggest open source models at home takes eight to ten thousand dollars of compute, so a hundred-dollar subscription is cheaper for most people. Free code-named models train on your data, so keep them for side projects.&lt;/p&gt;&lt;p&gt;Ryan Doser had Kimi K3, GLM 5.3 Flash, and GLM 5.3 redesign his opt-in page through OpenRouter. Aaron said the best one looked like Opus 4.8, and that the best Kimi and GLM models are closer to the frontier than a year ago.&lt;/p&gt;&lt;h2&gt;Questions&lt;/h2&gt;&lt;h3&gt;Can you run an open source AI model on your phone without internet?&lt;/h3&gt;&lt;p&gt;Yes. Aaron uses Google AI Edge Gallery, a free app in the iPhone App Store that&amp;apos;s also on Android. It downloads Gemma models of about two and a half to five gigabytes, shows which ones fit on your phone, and runs them with Wi-Fi off, including image input. He uses it on airplanes and for medical, banking, and personal questions he doesn&amp;apos;t want leaving his device.&lt;/p&gt;&lt;h3&gt;Is it cheaper to host a big open source model yourself?&lt;/h3&gt;&lt;p&gt;Not for most people. Aaron says running a model like Kimi or GLM 5.3 at home takes eight to ten thousand dollars of compute, while a hundred-dollar Claude or OpenAI subscription gets about the same usage most people need. Small local models, like the 3 GB Qwen model he uses to rename screenshots, are where self-hosting pays off.&lt;/p&gt;&lt;h3&gt;Why use OpenRouter instead of separate API keys?&lt;/h3&gt;&lt;p&gt;Aaron uses one OpenRouter key instead of storing and swapping a key for every provider in his apps and environment files. It also lets him try a new model on a five-minute project for a dollar or fifty cents instead of paying 20 to 100 dollars a month for another plan.&lt;/p&gt;&lt;h3&gt;Where can you use open source models if Claude Code and the Codex app only run their own?&lt;/h3&gt;&lt;p&gt;Aaron uses the Hermes desktop app from Nous Research. Its model picker holds his ChatGPT subscription models, his Z.ai GLM plan, MiniMax, and the Nous Portal&amp;apos;s list of models, and it can send one prompt to DeepSeek, Kimi, and OpenAI at once to compare the answers side by side.&lt;/p&gt;&lt;h3&gt;Which open source models does Aaron use most?&lt;/h3&gt;&lt;p&gt;GLM Flash models from Z.ai, for at least nine months, for background tasks like searches, finding files, drafts, his family calendar, and logging workouts from a photo. GLM 5.3 Flash is the first one with image input. For price, he points to DeepSeek. He keeps free code-named models for side projects, because they train on everything you send.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>I edit most of my videos with Claude in a terminal now</title>
    <link>https://aaronmakelky.com/blog/ai-video-editing-with-claude</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/ai-video-editing-with-claude</guid>
    <pubDate>Sat, 10 Oct 2026 12:00:00 GMT</pubDate>
    <description>Ryan Doser and I talked AI video editing on AI Rabbit Holes. I still record in Riverside and Descript, but Claude, HyperFrames, and Tesseract do most of my editing now.</description>
    <content:encoded>&lt;h2&gt;How Aaron edits video with AI&lt;/h2&gt;&lt;p&gt;Aaron Makelky records multi-person episodes in Riverside or Descript for separate tracks and high-quality audio, then does most of his editing by pointing Claude Code or Codex at the raw file in a terminal.&lt;/p&gt;&lt;p&gt;His most used tool is HyperFrames, a free HTML-based video tool from HeyGen. He also uses Tesseract, and DaVinci Resolve with scripts Claude writes. None of them record, and some require a Mac and a lot of RAM.&lt;/p&gt;&lt;p&gt;For AI Rabbit Holes, Aaron connects Claude Code to the Riverside MCP and has a clipping skill find ten clips from the transcript. His caption rules keep text below his chin, never over his face, with tool logos and the main speaker on top in two-person clips.&lt;/p&gt;&lt;h2&gt;What he recommends&lt;/h2&gt;&lt;p&gt;Save the editing instructions as a skill, add one new thing each episode, and make it one to five percent better every time.&lt;/p&gt;&lt;p&gt;Use generated media like B-roll on top of real footage instead of generating the whole video, and turn finished edits into how-to posts.&lt;/p&gt;&lt;p&gt;Ryan Doser runs the podcast side with Claude Code skills for cuts, a Seedance-powered intro, packaging, and YouTube-to-blog repurposing.&lt;/p&gt;&lt;h2&gt;Questions&lt;/h2&gt;&lt;h3&gt;Can Claude edit a video without video editing software?&lt;/h3&gt;&lt;p&gt;Yes. Aaron points Claude Code or Codex at a raw video saved on his computer and has it edit with HyperFrames, which is free from HeyGen, or Tesseract. Claude writes the code for captions, graphics, explainer animations, and cuts in a terminal, so there&amp;apos;s no editing app or subscription for it to connect to.&lt;/p&gt;&lt;h3&gt;Do I still need Riverside or Descript if Claude does the editing?&lt;/h3&gt;&lt;p&gt;For recording, yes. HyperFrames and Tesseract don&amp;apos;t record. Aaron records with Riverside or Descript to get separate tracks and high-quality audio, especially with more than one person, then does most of the editing with Claude.&lt;/p&gt;&lt;h3&gt;What computer do I need for AI video editing?&lt;/h3&gt;&lt;p&gt;Something with local access to the file and powerful hardware. Some of these tools require a Mac and don&amp;apos;t run on Linux, so a cloud server may not work, and editing uses a lot of RAM. Aaron says a MacBook Air or a Chromebook isn&amp;apos;t enough.&lt;/p&gt;&lt;h3&gt;How do I get better results from AI video edits?&lt;/h3&gt;&lt;p&gt;Aaron saves his editing instructions as a skill, including a brand guide for captions, logos, and two-person layouts. He uses it on every episode, tries one new thing each time, and records what to keep doing and what never to do again.&lt;/p&gt;&lt;h3&gt;Should I generate the whole video with AI?&lt;/h3&gt;&lt;p&gt;Aaron doesn&amp;apos;t. He uses generated media like B-roll that demonstrates what he&amp;apos;s talking about, while the main footage is still him on camera. He expects platforms to keep limiting how fully generated videos are monetized and labeled.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>I made a nature documentary about my gym fridge</title>
    <link>https://aaronmakelky.com/blog/liquid-death-nature-documentary-edit</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/liquid-death-nature-documentary-edit</guid>
    <pubDate>Fri, 09 Oct 2026 12:00:00 GMT</pubDate>
    <description>I filmed it in 33 minutes and spent the evening editing it with Claude Code, HeyGen, and FFmpeg. Here&apos;s every prompt I sent and what I&apos;d do differently next time.</description>
    <content:encoded>&lt;h2&gt;How the video was made&lt;/h2&gt;&lt;p&gt;Aaron filmed 23 silent clips of stocking his home-gym mini fridge with Liquid Death on an Insta360 Luna Ultra: mostly 4K at 30 fps, with the walk-in at 1080p 240 fps and the pink bottle at 4K 120 fps.&lt;/p&gt;&lt;p&gt;Claude Code on Claude Opus 5.5 in the Claude desktop app copied the footage, matched it to a 14-beat Notion shot list, cut it with FFmpeg and Python, and wrote a deadpan nature-documentary script that Aaron approved before any audio was generated.&lt;/p&gt;&lt;p&gt;HeyGen text to speech voiced 14 lines with a voice Aaron designed in HeyGen, for $0.23 total. The word timestamps from HeyGen drove the captions.&lt;/p&gt;&lt;h2&gt;What the final edit does&lt;/h2&gt;&lt;p&gt;It ships as a 16:9 master and two verticals: the full frame over a blurred copy of itself with captions in the blurred band, and a native 9:16 crop.&lt;/p&gt;&lt;p&gt;The captions use Gill Sans with the active word inverted, and every caption state sits inside a 1080 by 1920 safe box of x 60 to 940 and y 250 to 1500.&lt;/p&gt;&lt;p&gt;The video opens on a real frame instead of black, loops with a last-to-first frame similarity of 0.98, keeps music under the narration without ever dropping to silence, and ends on a push-in to the Liquid Death sticker.&lt;/p&gt;&lt;h2&gt;Questions&lt;/h2&gt;&lt;h3&gt;What camera shot the Liquid Death fridge video?&lt;/h3&gt;&lt;p&gt;An Insta360 Luna Ultra. 23 clips, mostly 4K at 30 fps, with the walk-in at 1080p 240 fps and the pink bottle at 4K 120 fps.&lt;/p&gt;&lt;h3&gt;How was the narration made?&lt;/h3&gt;&lt;p&gt;HeyGen text to speech with a voice Aaron designed in HeyGen, generated one line at a time through the HeyGen CLI after he approved the script. All 14 lines cost $0.23.&lt;/p&gt;&lt;h3&gt;Is it a David Attenborough voice?&lt;/h3&gt;&lt;p&gt;No. The ticket ruled out any Attenborough clone or soundalike. It&amp;apos;s a generic deadpan narrator.&lt;/p&gt;&lt;h3&gt;How do you get a 16:9 video into a 9:16 vertical without cropping?&lt;/h3&gt;&lt;p&gt;Place the full 16:9 frame in the middle of the vertical frame over a blurred, darkened copy of itself. The blurred band above it is room for captions.&lt;/p&gt;&lt;h3&gt;How does the video loop seamlessly?&lt;/h3&gt;&lt;p&gt;The video starts 1.5 seconds into the opening shot and ends with that shot&amp;apos;s first 1.5 seconds, so the last frame runs into frame 1. The music does the same trick with the track&amp;apos;s first 1.5 seconds.&lt;/p&gt;&lt;h3&gt;Was this sponsored?&lt;/h3&gt;&lt;p&gt;It&amp;apos;s part of Aaron&amp;apos;s Liquid Death partnership. Liquid Death sent him free Liquid Death for a year.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>I one-shot a CarPlay video edit with Claude and Tesseract</title>
    <link>https://aaronmakelky.com/blog/one-shot-video-edit-claude-tesseract</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/one-shot-video-edit-claude-tesseract</guid>
    <pubDate>Sun, 04 Oct 2026 12:00:00 GMT</pubDate>
    <description>I sent one prompt and gave zero revision notes. Here are the real prompts, Linear ticket, brand rules, and outputs behind a 2:11 ChatGPT dot vs Grok Bot CarPlay video, and how to set up your own version.</description>
    <content:encoded>&lt;h2&gt;How the video was made&lt;/h2&gt;&lt;p&gt;Aaron recorded a 2 minute 52 second 4K take on an Insta360 Luna Ultra with an Insta360 Mic Pro, calling ChatGPT dot and Grok Bot over Apple CarPlay.&lt;/p&gt;&lt;p&gt;He sent Claude Code, running Claude Opus 5.5 at medium effort, one prompt asking for a clear, on-brand edit that never covers his face or a screen he is pointing at, tracked in Linear.&lt;/p&gt;&lt;p&gt;A cloud session wrote the brief into Linear ticket AAR-270. A local Claude Code session on his Mac Studio picked up the ticket, edited the video in Tesseract with his video-edit-tesseract skill, and returned a 2:11 edit that Aaron posted with no revision notes.&lt;/p&gt;&lt;h2&gt;What made it one-shottable&lt;/h2&gt;&lt;p&gt;A brand system written as checkable rules: colors with jobs, two font files, motion limits, and places graphics can never go.&lt;/p&gt;&lt;p&gt;A saved editing skill, a ticket that defined deliverables before work started, and footage shot with clean audio and room for captions.&lt;/p&gt;&lt;p&gt;The deliverables included an editable Tesseract project, a review MP4, SRT and VTT captions, an edit decision list with 30 source timecodes, a post caption, and verification checks including OCR for private text.&lt;/p&gt;&lt;h2&gt;Questions&lt;/h2&gt;&lt;h3&gt;What does one-shot mean for this video edit?&lt;/h3&gt;&lt;p&gt;Aaron sent one creative prompt and approved the first rendered version with no revision notes. Two later messages only told Claude where the source file lived and handed the Linear ticket to a local session.&lt;/p&gt;&lt;h3&gt;What tools were used to edit the video?&lt;/h3&gt;&lt;p&gt;Claude Code running Claude Opus 5.5 at medium effort, Tesseract with Aaron&amp;apos;s video-edit-tesseract skill, Linear for tracking, Claude Remote Control to follow the local session, Whisper and forced alignment for word timing, and FFmpeg for checks.&lt;/p&gt;&lt;h3&gt;What camera and microphone recorded the source?&lt;/h3&gt;&lt;p&gt;An Insta360 Luna Ultra recording 4K at 24 fps in a wide 16:9 frame, with audio from an Insta360 Mic Pro, the version with the e-ink display.&lt;/p&gt;&lt;h3&gt;Do I need Aaron&amp;apos;s brand system to get the same result?&lt;/h3&gt;&lt;p&gt;No. You need your own written brand rules: a palette with a job for each color, two font files, motion limits, and places graphics are never allowed to go.&lt;/p&gt;&lt;h3&gt;Why did the edit include SRT and VTT caption files?&lt;/h3&gt;&lt;p&gt;The Linear ticket listed them as deliverables before any editing started.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>How to make your own ChatGPT dots in Grok Bot, Claude, or Muse</title>
    <link>https://aaronmakelky.com/blog/chatgpt-dots-in-grok-bot</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/chatgpt-dots-in-grok-bot</guid>
    <pubDate>Wed, 30 Sep 2026 12:00:00 GMT</pubDate>
    <description>ChatGPT&apos;s dots run a background Heartbeat Dreamer that starts at a $100-a-month plan. Here&apos;s the same prompt and a setup for each agent: Grok Bot routines, Claude scheduled tasks, or Muse recurring tasks.</description>
    <content:encoded>&lt;h2&gt;Article&lt;/h2&gt;&lt;p&gt;ChatGPT&amp;apos;s new dots are always-on agents that keep working after you close the chat. Getting one takes a ChatGPT plan that starts at $100 a month. I set mine up in the ChatGPT desktop app this morning, and it spawned a background agent called &amp;quot;Heartbeat Dreamer.&amp;quot; Its full prompt is at the bottom of this page.&lt;/p&gt;&lt;p&gt;You can build your own version of that system without the $100 plan. This guide covers three agents: Grok Bot , Claude , and Muse . The parts that work the same everywhere come first, then each agent gets its own setup steps, prompts, and limits. If you already run a chief of staff bot and project leads in Grok Bot, this fits that org chart: a green specialist that researches on a schedule and hands findings to your chief.&lt;/p&gt;&lt;h2&gt;What it does&lt;/h2&gt;&lt;p&gt;OpenAI calls this proactive research. When you aren&amp;apos;t talking to your dot, it looks for ways to help in the background, and OpenAI says those background tools are read-only: no sending messages, no editing your apps, no driving your browser. ( OpenAI )&lt;/p&gt;&lt;p&gt;The prompt shows how that&amp;apos;s wired up. The dot you chat with is the &amp;quot;parent.&amp;quot; The Heartbeat Dreamer is a child agent that only researches, and it reports to the parent. It&amp;apos;s allowed to pass along private or sensitive stuff, and the parent decides what reaches you.&lt;/p&gt;&lt;p&gt;They talk through files. The parent keeps /user_notes/ and /action_items.md. The dreamer keeps its own checkpoints in /agent_notes/. It never edits the parent&amp;apos;s notes. When something should change, it tells the parent to change it.&lt;/p&gt;&lt;p&gt;Every run starts by reading the last checkpoint, the parent&amp;apos;s notes, roughly the last two hours of your chat with the parent, and a shared notes page if you have one. Then it makes a short plan and goes through your connected apps, email first. It ends by writing a new checkpoint that says where the next run should pick up, and it deletes its own checkpoints once they&amp;apos;re more than 14 days old.&lt;/p&gt;&lt;p&gt;The last step is one message to the parent, with every finding tagged P00, P0, P1 and so on. If nothing is worth sending, it stays quiet and writes down why.&lt;/p&gt;&lt;p&gt;Once a day, around 3am your time, it does a reflection run instead. That run rereads the day&amp;apos;s conversation, picks up preferences and any offers you turned down, and reshuffles your priorities.&lt;/p&gt;&lt;h2&gt;The rules to copy&lt;/h2&gt;&lt;p&gt;These are the lines you can drop straight into your agents:&lt;/p&gt;&lt;h2&gt;How to build it in Grok Bot&lt;/h2&gt;&lt;p&gt;The Grok Bot version is a routine on a specialist Bot that reads your connectors and browser sessions on a timer, writes checkpoints to a shared file on the cloud computer, and messages your chief Bot with tagged findings. Routines run in the cloud while your laptop is closed. ( xAI: Skills and routines ) Grok Bot access requires a paid individual Cursor plan, a Cursor Teams seat, or a linked SuperGrok subscription. ( xAI: Get started )&lt;/p&gt;&lt;p&gt;Here&amp;apos;s how you set it up.&lt;/p&gt;&lt;p&gt;Heartbeat prompt (swap in your chief Bot&amp;apos;s @ name, connectors, and time zone):&lt;/p&gt;&lt;p&gt;Create a routine called &amp;quot;Heartbeat&amp;quot; on this Bot that runs on weekdays at 7:00 AM, 10:00 AM, 1:00 PM, and 4:00 PM in America/Denver. Each run should: 1. Read /workspace/heartbeat/notes.md: the latest entry under Checkpoints and anything under Inbox I have not acted on. 2. Research my connected sources, email first, then calendar, then any connectors I have enabled (use @ in this chat to reach them). Use browser sessions on the shared computer where needed. Look at what changed since the last checkpoint: new threads, replies, invites, deadlines, things I said I would do, and things I was asked to do. 3. Before calling anything unanswered, overdue, or urgent, check sent mail, later replies, and task status. A search date range is not proof you reviewed that whole range. 4. Stay read-only. Do not send, reply, accept, edit, or delete anything in any app without my approval. The only file you may write is /workspace/heartbeat/notes.md. 5. Add a checkpoint under Checkpoints with the run time, what you checked, what you could not reach, and where the next run should pick up. 6. Add findings under Inbox. Tag each one P0, P1, or P2, and include what you learned, how sure you are, the source, and one concrete next step. If nothing is worth adding, write &amp;quot;Quiet&amp;quot; and one line on why. 7. Send an async message to @Chief with the same tagged findings. If nothing needs the chief, stay quiet in chat and record why in the checkpoint. 8. Treat a one-time choice as a one-time choice. If the same blocker shows up across several runs, ask me once whether to pause this routine. Reflection prompt:&lt;/p&gt;&lt;p&gt;Create a routine called &amp;quot;Heartbeat reflection&amp;quot; on this Bot that runs every day at 2:45 AM America/Denver. Each run should: 1. Read /workspace/heartbeat/notes.md and, if you can reach them, my recent conversations with my chief Bot from the past day. 2. Add a checkpoint under Checkpoints titled Reflection: preferences I stated, offers I turned down, and whether the open Inbox items are ranked right. Keep one-off comments as one-offs. 3. Stay read-only everywhere except /workspace/heartbeat/notes.md. 4. Message @Chief only if reflection changes how it should prioritize my Inbox. Where the Grok Bot version falls short Dot&amp;apos;s read-only rule is enforced in code. In Grok Bot it is a boundary in the Bot description, the routine instructions, and optional Auto-review rules. Connectors and browser tools can still attempt writes unless you stop them. ( xAI: Approvals )&lt;/p&gt;&lt;p&gt;There is no built-in /agent_notes/ path. You use a file under /workspace and Bot messages instead. That file is visible to every Bot on your account because they share one cloud computer. ( xAI: Shared computer )&lt;/p&gt;&lt;p&gt;The dreamer reads the last two hours of parent chat every run in OpenAI&amp;apos;s prompt. Grok Bot docs describe Bot memory, search, and handoffs, but not an automatic two-hour parent transcript on every routine run. Your chief Bot has to pull context from conversation history, your messages, or the notes file.&lt;/p&gt;&lt;p&gt;Routines draw on your Grok Bot usage, and xAI says it may pause routines to control unattended usage. Heavy schedules plus connector research can burn your weekly limit fast. ( xAI: Skills and routines , how I stretch Grok Bot usage )&lt;/p&gt;&lt;p&gt;xAI&amp;apos;s Grok Bot docs do not publish a full Marketplace plugin list or confirm every connector Claude lists (GitHub, Linear, Notion, and so on). Install what you actually use and record gaps in your Bot description rather than assuming parity.&lt;/p&gt;&lt;h2&gt;How to build it in Claude&lt;/h2&gt;&lt;p&gt;The Claude version is a scheduled task that reads your apps on a timer and writes to one notes doc, plus a second task at night for the reflection. Scheduled tasks are on Claude&amp;apos;s paid plans (Pro, Max, Team, Enterprise) and run in the cloud, so they keep going when your laptop is asleep. ( Claude Help Center )&lt;/p&gt;&lt;p&gt;Here&amp;apos;s how you set it up.&lt;/p&gt;&lt;p&gt;Heartbeat prompt (swap in your own times and time zone):&lt;/p&gt;&lt;p&gt;Create a scheduled task called &amp;quot;Heartbeat&amp;quot; that runs on weekdays at 7am, 10am, 1pm, and 4pm Mountain time. Each run should: 1. Read my Google Doc &amp;quot;Heartbeat notes&amp;quot;: the latest entry under Checkpoints, and anything under Inbox I haven&amp;apos;t acted on. 2. Research my connected apps, email first, then calendar, then everything else. Look at what changed since the last checkpoint: new threads, replies, invites, deadlines, things I said I&amp;apos;d do, and things I was asked to do. 3. Before calling anything unanswered, overdue, or urgent, check my sent mail, later replies, and task status. A search&amp;apos;s date range is not proof you reviewed that whole range. 4. Stay read-only. Do not send, reply, accept, edit, or delete anything in any app. The only thing you may write to is the Heartbeat notes doc. 5. Add a checkpoint under Checkpoints: the run time, what you checked, what you couldn&amp;apos;t reach, and where the next run should pick up. 6. Add findings under Inbox. Tag each one P0, P1, or P2, and include what you learned, how sure you are, the source, and one concrete next step. If nothing is worth adding, write &amp;quot;Quiet&amp;quot; and one line on why. 7. Treat a one-time choice as a one-time choice. If the same blocker shows up across several runs, ask me once whether to pause this task. Reflection prompt:&lt;/p&gt;&lt;p&gt;Create a scheduled task called &amp;quot;Heartbeat reflection&amp;quot; that runs every day at 2:45am Mountain time. Each run should: 1. Read my Google Doc &amp;quot;Heartbeat notes&amp;quot; and, if you can reach them, my Claude chats from the past day. 2. Add an entry under Checkpoints titled Reflection: preferences I stated, offers I turned down, and whether the open Inbox items are ranked right. Keep one-off comments as one-offs. 3. Stay read-only everywhere except the Heartbeat notes doc. Where the Claude version falls short Dot&amp;apos;s read-only rule is enforced in code. In this setup it&amp;apos;s a sentence in a prompt. If you want a harder stop, set the task up manually from Scheduled and pick an approval mode that makes it ask before it acts.&lt;/p&gt;&lt;p&gt;There&amp;apos;s no parent agent. Dot&amp;apos;s dreamer hands findings to the agent you already talk to. Here, findings sit in a doc until you, or your main chat, go read them.&lt;/p&gt;&lt;p&gt;It doesn&amp;apos;t read your conversations every run. The dreamer checks the last two hours of your chat with the parent each time. The notes doc in step 3 is how you cover that.&lt;/p&gt;&lt;h2&gt;How to build it in Muse&lt;/h2&gt;&lt;p&gt;The Muse version is a recurring scheduled task that reads your Connectors on a timer and writes to one notes file in your Library, plus a second task at night for reflection. Meta says recurring tasks repeat on the schedule you set and are delivered as messages in your Muse conversation. ( Meta Help Center: reminders and scheduled tasks )&lt;/p&gt;&lt;p&gt;Here&amp;apos;s how you set it up.&lt;/p&gt;&lt;p&gt;Heartbeat prompt (swap in your own times and time zone):&lt;/p&gt;&lt;p&gt;Create a recurring scheduled task called &amp;quot;Heartbeat&amp;quot; that runs on weekdays at 7am, 10am, 1pm, and 4pm Mountain time. Each run should: 1. Read my Library file &amp;quot;Heartbeat notes&amp;quot;: the latest entry under Checkpoints, and anything under Inbox I have not acted on. 2. Research my connected apps through Connectors, email first, then calendar, then anything else I have connected. Look at what changed since the last checkpoint: new threads, replies, invites, deadlines, things I said I would do, and things I was asked to do. 3. Before calling anything unanswered, overdue, or urgent, check my sent mail, later replies, and task status. A search&amp;apos;s date range is not proof you reviewed that whole range. 4. Stay read-only on Connectors. Do not send, reply, accept, edit, or delete anything in any connected app. The only thing you may write to is the Heartbeat notes file in my Library. 5. Add a checkpoint under Checkpoints: the run time, what you checked, what you could not reach, and where the next run should pick up. 6. Add findings under Inbox. Tag each one P0, P1, or P2, and include what you learned, how sure you are, the source, and one concrete next step. If nothing is worth adding, write &amp;quot;Quiet&amp;quot; and one line on why. 7. Treat a one-time choice as a one-time choice. If the same blocker shows up across several runs, ask me once whether to pause this task. 8. Deliver a short summary in this chat when the run finishes. Reflection prompt:&lt;/p&gt;&lt;p&gt;Create a recurring scheduled task called &amp;quot;Heartbeat reflection&amp;quot; that runs every day at 2:45am Mountain time. Each run should: 1. Read my Library file &amp;quot;Heartbeat notes&amp;quot; and, if you can reach them, my Muse chats from the past day. 2. Add an entry under Checkpoints titled Reflection: preferences I stated, offers I turned down, and whether the open Inbox items are ranked right. Keep one-off comments as one-offs. 3. Stay read-only on Connectors. The only file you may edit is Heartbeat notes in my Library. 4. Deliver a short summary in this chat when the run finishes. Where the Muse version falls short Dot&amp;apos;s read-only rule is enforced in code. In Muse it is a mix of prompt instructions, Connector retrieve-only settings, and approval prompts. Meta says many approval and security checks run outside the AI model, but this is not the same as a dedicated background researcher with hard-coded read-only tools. ( Meta Help Center: privacy and security )&lt;/p&gt;&lt;p&gt;There is no parent agent. Dot&amp;apos;s dreamer hands findings to the agent you already talk to. Findings land in a Library file and in scheduled-task messages until you open your main chat and ask for them.&lt;/p&gt;&lt;p&gt;Meta&amp;apos;s help articles do not say each scheduled run automatically reads the last two hours of your main chat. The Heartbeat notes file is the shared handoff.&lt;/p&gt;&lt;p&gt;Meta&amp;apos;s Connectors article doesn&amp;apos;t name project tools like Linear or Notion. Services outside the in-app list require a Custom Connector, which Meta does not review. ( Meta Help Center: Connectors )&lt;/p&gt;&lt;p&gt;Recurring tasks can repeat on a custom interval, but Meta does not document a maximum cadence or how many tokens a research-heavy schedule burns on the free tier. If heartbeats hit your usage limit, you wait for the refresh or upgrade. Power is $20 a month for a higher weekly allowance. ( Meta Help Center: Muse subscriptions )&lt;/p&gt;&lt;h2&gt;The full prompt&lt;/h2&gt;&lt;p&gt;OpenAI wrote this prompt, not me, and it will probably change. Raw text file&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>I asked an AI agent to edit my video. Here are all three versions.</title>
    <link>https://aaronmakelky.com/blog/tesseract-video-edit-process</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/tesseract-video-edit-process</guid>
    <pubDate>Mon, 21 Sep 2026 12:00:00 GMT</pubDate>
    <description>My exact prompts, three before-and-after edits, and the caption, sound, and camera rules I saved for the next Tesseract video.</description>
    <content:encoded>&lt;h2&gt;One take, three edits&lt;/h2&gt;&lt;p&gt;The 57.375-second original became a 54.75-second comparison. Version 2 improved caption placement, animated graphics, the coffee cue, and the ending. Version 3 refined 42 caption phrases and varied nine sound cues.&lt;/p&gt;&lt;h2&gt;The next edit starts with saved instructions&lt;/h2&gt;&lt;p&gt;The custom video-edit-tesseract skill records the approved video style. Subtle camera movement that preserves eye height was added to the skill after version 3 and has not been rendered into that video.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>Meow vs Mercury: why my business banking went to Mercury</title>
    <link>https://aaronmakelky.com/blog/meow-vs-mercury-api-comparison</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/meow-vs-mercury-api-comparison</guid>
    <pubDate>Fri, 18 Sep 2026 12:00:00 GMT</pubDate>
    <description>I compared Meow and Mercury for my consulting business: the apps, the APIs, the pricing, and what my agents can actually do with each one. Mercury won as the daily operating account; Meow is still the better idea for how agents should touch money.</description>
    <content:encoded>&lt;h2&gt;Article&lt;/h2&gt;&lt;p&gt;I needed a real business bank for my consulting practice. A fractional CFO friend recommended Meow, then admitted he banks at Mercury. Wells Fargo was the incumbent I was trying to escape. So I compared Meow and Mercury properly: the apps, the APIs, the pricing, and what my agents can actually do with each one.&lt;/p&gt;&lt;p&gt;Mercury won as my daily operating account. Meow won as the better idea for how agents should touch money.&lt;/p&gt;&lt;h2&gt;Mercury is the better daily driver.&lt;/h2&gt;&lt;p&gt;For a solo consultant prioritizing mobile banking, invoicing, webhooks, and automation , Mercury is the more complete operating setup.&lt;/p&gt;&lt;p&gt;Best overall fit · with a $29.90/mo caveat CHANGE THE DECISION LENS Run the business Let agents take action Avoid monthly fees MERCURY Native iOS and Android apps, mature receivables, approval alerts, and broad banking APIs make it the stronger everyday operating account. MEOW Its official MCP can create products, customers, and invoices, manage virtual cards, and draft transfers for human approval. Mercury’s official MCP is read-only. MEOW Meow has no invoicing-API paywall. Mercury requires Plus at $29.90 per month for Invoices and Customers API access. THE SHORT VERSION Choose the workflow, not just the API. Mercury wins the human layer. Its native mobile apps handle balances, transfers, cards, mobile check deposit, receipts, and pending-approval alerts. Meow wins the agent layer: its official MCP can create invoices, issue or freeze virtual cards, and draft transfers for human approval.&lt;/p&gt;&lt;p&gt;Pick Mercury if the bank account needs to work well from your phone.&lt;/p&gt;&lt;p&gt;Pick Meow if coding-agent write access matters more than a mobile app.&lt;/p&gt;&lt;h2&gt;Mobile app&lt;/h2&gt;&lt;p&gt;Native iOS + Android — Meow has no native app.&lt;/p&gt;&lt;h2&gt;Webhooks&lt;/h2&gt;&lt;p&gt;Signed delivery, retries, and event filtering on both sides.&lt;/p&gt;&lt;h2&gt;Invoicing API&lt;/h2&gt;&lt;p&gt;Both create and track invoices; Mercury requires Plus.&lt;/p&gt;&lt;h2&gt;Coding agents&lt;/h2&gt;&lt;p&gt;Official MCP writes invoices and drafts transfers.&lt;/p&gt;&lt;h2&gt;What changes in practice&lt;/h2&gt;&lt;p&gt;The two APIs overlap heavily. The decisive differences are mobile access, MCP write capability, and the friction around invoicing and write tokens.&lt;/p&gt;&lt;p&gt;Meow explicitly positions banking as shifting away from apps toward AI agents.&lt;/p&gt;&lt;p&gt;Balances, transactions, transfers, mobile check deposit, cards, receipt scanning, widgets, and push alerts for pending approvals.&lt;/p&gt;&lt;p&gt;Products, customers, invoices, line items, PDF downloads, recipient email, multiple payment methods, and lifecycle status.&lt;/p&gt;&lt;p&gt;Create, send, update, cancel, track, and download PDFs. Requires Mercury Plus at $29.90/month or higher.&lt;/p&gt;&lt;p&gt;Create products, customers, and invoices; issue or manage virtual cards; draft transfers that stay pending until a person approves them.&lt;/p&gt;&lt;p&gt;Official tools list and inspect accounts, transactions, cards, invoices, customers, and more—but do not create invoices or send money.&lt;/p&gt;&lt;p&gt;Full read/write API with about 30 fine-grained scopes. No IP-whitelist requirement.&lt;/p&gt;&lt;p&gt;Read &amp;amp; Write tokens require IP whitelisting. Custom tokens can expose selected endpoints, but write scopes generally still need a whitelist.&lt;/p&gt;&lt;p&gt;HMAC-SHA-256, snapshot or thin payloads, test events, delivery history, redrive, secret rotation, and circuit breaking.&lt;/p&gt;&lt;p&gt;HMAC-SHA-256, event-type and path filters, test verification, and at-least-once delivery. Not available in sandbox.&lt;/p&gt;&lt;p&gt;Simulate inbound transfers, stablecoin funding, card authorization and settlement events, and application approval.&lt;/p&gt;&lt;p&gt;Preloaded data and sandbox-only tokens. Use the Events API because webhooks do not run in the sandbox.&lt;/p&gt;&lt;h2&gt;Drill into the details&lt;/h2&gt;&lt;p&gt;Open any section for the parts most likely to affect an integration plan.&lt;/p&gt;&lt;p&gt;Accounts, payments, cards &amp;amp; operations Meow Accounts, balances, transactions, limits, statements, tax documents, contacts, users, reimbursements, and routing-number validation. ACH, scheduled ACH, domestic wire, internal book, international, crypto, and USDC transfer creation. Create, update, and revoke cards; bill-pay CRUD and document upload flows. Business onboarding, representatives, KYC, due diligence, and document uploads are API-accessible. Mercury Accounts, transactions, statements, recipients, merchants, categories, credit, treasury, users, organization, and SAFEs. Send money to recipients, internal transfers, checks, and domestic wires; manage send-money approval requests. Create, update, cancel, freeze, and unfreeze cards; Vault can reveal card details. Attachments, transaction metadata, statement PDFs, and OAuth2 integration flows. Authentication &amp;amp; token management Meow API key in the x-api-key header, created and managed in the dashboard. Fine-grained scopes include invoices, products, customers, transfers, cards, webhooks, bill pay, and simulations. Multi-entity keys add an x-entity-id header. OAuth is reserved for the MCP surface. Mercury Bearer or basic authentication with Read Only, Read and Write, or Custom token tiers. Write access generally requires IP whitelisting; scopes cannot be edited after creation. Unused tokens are auto-deleted after 45 days; unused permissions can also be auto-downgraded after 45 days, with seven days’ email notice. Invoice lifecycle &amp;amp; payment methods Meow Status covers draft, open, pending, scheduled, partially paid, canceled, expired, and overdue. Tracks total, paid, and due amounts; supports invoice dates, due dates, notes, line items, recipient email, and PDF download. Documented payment methods include bank transfer, card, ACH debit, international wire, and several USDC/USDT networks. Mercury Create, list, get, update, cancel, attach files, download PDF, and send by email at creation. Customer payment methods include credit card, Apple Pay, Google Pay, wire, ACH, and ACH debit for subscribers. Plus adds recurring invoices and ACH-debit invoicing at $1 per transaction. IMPLEMENTATION PATH What each choice commits you to RECOMMENDED Mercury as the operating account Best if you want dependable phone access first, then layer automation around it.&lt;/p&gt;&lt;h2&gt;Meow as the automation account&lt;/h2&gt;&lt;p&gt;Best if you want a coding agent to do more through an official integration.&lt;/p&gt;&lt;h2&gt;What I actually need from a business bank&lt;/h2&gt;&lt;p&gt;I need five things: a mobile app I can run the business from, invoicing I can trigger from my own systems, an API my automations can use, transfers I can track, and a bank that doesn&amp;apos;t make my wife drive to a branch to get added to an account.&lt;/p&gt;&lt;h2&gt;The branch visit that ended the debate&lt;/h2&gt;&lt;p&gt;Wells Fargo made us go in person to get my wife added to a business account. In 2026. Mercury let me add her online, is shipping her a physical card, and sent 25 checks for free.&lt;/p&gt;&lt;p&gt;Wells Fargo treated its process as sacred and my time as worthless. Mercury acted like software. I know which one I want holding my operating cash.&lt;/p&gt;&lt;h2&gt;Where Mercury wins: the daily operating account&lt;/h2&gt;&lt;p&gt;Mercury is the better bank for the unglamorous daily work. The mobile app is native and actually good, which matters because I run most of my business from my phone. Transfers have real tracking instead of the &amp;quot;it&amp;apos;ll show up in 1-3 business days, good luck&amp;quot; experience. Webhooks mean my systems know when money moves without me polling an API like it&amp;apos;s 2014.&lt;/p&gt;&lt;p&gt;The invoicing API is the sleeper feature. I can generate and send invoices from my own workflows instead of living inside someone else&amp;apos;s dashboard. It requires a Plus plan, which is worth knowing before you build on it.&lt;/p&gt;&lt;p&gt;Mercury&amp;apos;s API story has one real wart for my use case: general read and write tokens require IP whitelisting. That&amp;apos;s fine for a server with a static IP and annoying for agents running on dynamic infrastructure. Their official MCP was documented as beta and read-only, so don&amp;apos;t plan on your agent doing much inside Mercury yet.&lt;/p&gt;&lt;h2&gt;Where Meow wins: agent-native money movement&lt;/h2&gt;&lt;p&gt;Meow is built around a genuinely better idea for how agents should touch money: the agent prepares transfers and a human approves them. That&amp;apos;s the workflow I actually want. I don&amp;apos;t want my agent moving money autonomously, and I don&amp;apos;t want to be the API either. Human-approved drafts are the setup I want for agent-run finance: the agent drafts, I approve.&lt;/p&gt;&lt;p&gt;The pricing is unusual: no public paid subscription for the core banking and agent offering. The paid parts are gated by balance instead of subscription. Yield and brokerage products need a $100,000 checking balance, and the treasury product runs 1 basis point per month, annualized.&lt;/p&gt;&lt;p&gt;The weak spot is mobile: Meow has no native mobile app, and I run my business from my phone. That is the gap that keeps it from being my daily driver.&lt;/p&gt;&lt;h2&gt;The pricing&lt;/h2&gt;&lt;p&gt;Mercury is straightforward: a free tier, Plus at $29.90 a month, Pro at $299 a month. The invoicing API needs Plus.&lt;/p&gt;&lt;p&gt;Meow has no public paid subscription for the core offering. The costs show up as balance gates and treasury fees instead. Different model, not necessarily a cheaper one, depending on how much cash you keep parked.&lt;/p&gt;&lt;h2&gt;My verdict&lt;/h2&gt;&lt;p&gt;Mercury is my operating account. It does the daily work better: mobile, invoices, webhooks, tracking, and a bank that treats my time like it matters. Meow is the one I&amp;apos;m watching. The human-approved transfer draft is how agent-run finance should work. If they ship a real mobile app and keep the agent-approval workflow ahead of Mercury&amp;apos;s, I&amp;apos;ll take another look.&lt;/p&gt;&lt;p&gt;If you&amp;apos;re picking today and you mostly need a bank that works, pick Mercury. If you&amp;apos;re building agent workflows that touch money and you want a human in the approval loop by design, Meow deserves a real look.&lt;/p&gt;&lt;p&gt;Disclosure: the Mercury link below is my referral link. It costs you nothing and supports the site. If you&amp;apos;d rather not use it, mercury.com works the same.&lt;/p&gt;&lt;p&gt;Open a Mercury account with my referral link&lt;/p&gt;&lt;h2&gt;Should I pick Meow or Mercury for my business banking?&lt;/h2&gt;&lt;p&gt;I picked Mercury as my daily operating account because of its mobile app, invoicing API, webhooks, and transfer tracking. Meow is the better pick if you want agent-native workflows with human-approved transfer drafts. If you mostly need a bank that just works day to day, I&amp;apos;d start with Mercury.&lt;/p&gt;&lt;h2&gt;How much does Mercury cost?&lt;/h2&gt;&lt;p&gt;Mercury has a free tier, Plus at $29.90 a month, Pro at $299 a month. The invoicing API requires a Plus plan or higher.&lt;/p&gt;&lt;h2&gt;Does Meow charge a subscription?&lt;/h2&gt;&lt;p&gt;No. Meow&amp;apos;s core banking and agent offering has no public paid subscription. The paid parts are gated by balance instead: yield and brokerage products require a $100,000 checking balance, and the treasury product charges 1 basis point per month, annualized.&lt;/p&gt;&lt;h2&gt;Can my AI agents move money with Mercury or Meow?&lt;/h2&gt;&lt;p&gt;Meow is built for this: agents can prepare transfers that a human approves. Mercury&amp;apos;s official MCP was documented as beta and read-only, and its general read/write API tokens require IP whitelisting, which is friction for agents running on dynamic infrastructure. For agent-driven money movement with a human in the loop, Meow has the better model right now.&lt;/p&gt;&lt;h2&gt;Did you really have to visit a branch to add your wife to Wells Fargo?&lt;/h2&gt;&lt;p&gt;Yes. Wells Fargo required an in-person visit to add my wife to a business account. Mercury let me add her online, shipped her a physical card, and sent 25 checks for free. That single experience told me more about the two banks than any feature table.&lt;/p&gt;&lt;h2&gt;Questions&lt;/h2&gt;&lt;h3&gt;Should I pick Meow or Mercury for my business banking?&lt;/h3&gt;&lt;p&gt;I picked Mercury as my daily operating account because of its mobile app, invoicing API, webhooks, and transfer tracking. Meow is the better pick if you want agent-native workflows with human-approved transfer drafts. If you mostly need a bank that just works day to day, I&amp;apos;d start with Mercury.&lt;/p&gt;&lt;h3&gt;How much does Mercury cost?&lt;/h3&gt;&lt;p&gt;Mercury has a free tier, Plus at $29.90 a month, Pro at $299 a month. The invoicing API requires a Plus plan or higher.&lt;/p&gt;&lt;h3&gt;Does Meow charge a subscription?&lt;/h3&gt;&lt;p&gt;No. Meow&amp;apos;s core banking and agent offering has no public paid subscription. The paid parts are gated by balance instead: yield and brokerage products require a $100,000 checking balance, and the treasury product charges 1 basis point per month, annualized.&lt;/p&gt;&lt;h3&gt;Can my AI agents move money with Mercury or Meow?&lt;/h3&gt;&lt;p&gt;Meow is built for this: agents can prepare transfers that a human approves. Mercury&amp;apos;s official MCP was documented as beta and read-only, and its general read/write API tokens require IP whitelisting, which is friction for agents running on dynamic infrastructure. For agent-driven money movement with a human in the loop, Meow has the better model right now.&lt;/p&gt;&lt;h3&gt;Did you really have to visit a branch to add your wife to Wells Fargo?&lt;/h3&gt;&lt;p&gt;Yes. Wells Fargo required an in-person visit to add my wife to a business account. Mercury let me add her online, shipped her a physical card, and sent 25 checks for free. That single experience told me more about the two banks than any feature table.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>Astra vs Fable: why I&apos;m still paying for ChatGPT</title>
    <link>https://aaronmakelky.com/blog/astra-vs-fable</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/astra-vs-fable</guid>
    <pubDate>Tue, 15 Sep 2026 12:00:00 GMT</pubDate>
    <description>Fable wrote a better draft from my 17-page PDF, but Astra finished faster. Ryan Doser and I compare writing, websites, and why I&apos;d choose ChatGPT Pro.</description>
    <content:encoded>&lt;h2&gt;Article&lt;/h2&gt;&lt;p&gt;I&amp;apos;ve used Fable 5.1 for long-form writing where I could immediately tell it did a better job than Astra, but that improvement hasn&amp;apos;t been enough for me to pay for a Claude subscription.&lt;/p&gt;&lt;p&gt;I gave both models the same PDF, 17 scanned pages of my handwritten ideas, and Fable had the better hook and did a better job organizing the writing, but Astra did the entire job faster than Fable could even read the PDF.&lt;/p&gt;&lt;p&gt;I prefer Claude Fable 5.1&amp;apos;s writing and still get more work done with GPT-6 Astra. How often do I need that extra bit of quality, and what am I giving up to get it?&lt;/p&gt;&lt;p&gt;I talked through that with Ryan Doser on AI Rabbit Holes on September 15, 2026. Ryan had been testing both models on websites, and I had been using them for writing and the other work I need to get done during the week.&lt;/p&gt;&lt;h2&gt;Where Fable 5.1 earned the extra time&lt;/h2&gt;&lt;p&gt;The writing task had a few places to go wrong before the model even got to the writing. It had to read a scanned PDF, pull the words out of my handwriting, and turn those ideas into a long-form piece without inventing a bunch of slop.&lt;/p&gt;&lt;p&gt;I gave Fable the PDF through Cursor and used the same prompt with Astra in ChatGPT, asking both to turn my handwritten ideas into a piece of writing that stayed true to what I&amp;apos;d actually said.&lt;/p&gt;&lt;p&gt;Fable took a long time and made a lot of tool calls just to read the PDF, but when I opened the result, I could see why I&amp;apos;d use it as a copywriting coach. I hadn&amp;apos;t given it a hook, just a pile of thoughts I wanted help structuring, and it did a much better job of figuring out how to put them together.&lt;/p&gt;&lt;p&gt;Even though I&amp;apos;m biased toward Astra because I pay for ChatGPT, I&amp;apos;d still pick Fable&amp;apos;s draft in a blind test.&lt;/p&gt;&lt;p&gt;Astra got through the PDF without getting stuck, and its writing had more personality than I&amp;apos;d come to expect from GPT. If I have one piece where I really need help with the hook and structure, I might reach for Fable, but if I&amp;apos;m doing ten of these in a day, the extra waiting becomes a much bigger part of the decision.&lt;/p&gt;&lt;h2&gt;Ryan&amp;apos;s landing page test: fancy doesn&amp;apos;t sell&lt;/h2&gt;&lt;p&gt;Ryan gave both models his existing landing page and asked for a better version designed to get more conversions. He hadn&amp;apos;t specifically told either model to stay on brand.&lt;/p&gt;&lt;p&gt;Fable stayed close to the page he already had, keeping the dark background and blue accents while making the purchase buttons stand out more. The testimonials also looked a little better to me, but overall it was an iteration of the existing design.&lt;/p&gt;&lt;p&gt;Astra changed the colors and put Ryan&amp;apos;s name at the top, then used a different widget to show what the product does. The page was shorter and more direct, which I actually liked even though it had gone off brand.&lt;/p&gt;&lt;p&gt;My gut reaction was that Astra&amp;apos;s version would probably perform better because it was easier to understand the offer and know where to click. Ryan would have to test that, but I thought the simpler version was worth putting in front of customers.&lt;/p&gt;&lt;p&gt;With a landing page, I want someone to understand what they&amp;apos;re buying and know where to click without having to work through a bunch of design. Ryan made the same point about a plumber in Des Moines: the business owner cares whether someone fills out the form and becomes a lead. A fancy page doesn&amp;apos;t help much if it gets in the way of that.&lt;/p&gt;&lt;h2&gt;Astra gave Ryan more of the business he asked for&lt;/h2&gt;&lt;p&gt;The second test started with a local service business idea: flooded-basement cleanup in Des Moines, Iowa. Ryan asked for a company name, an ideal customer profile, a brand guide, and a homepage design. He supplied a reference website and a skill.&lt;/p&gt;&lt;p&gt;Ryan had Claude Code and Codex running inside VS Code, so he could give them the same prompt from the workspace he already uses. He didn&amp;apos;t need to switch between two desktop apps to do the comparison.&lt;/p&gt;&lt;p&gt;Fable&amp;apos;s result looked rudimentary to me, with orange text that was hard to read and five steps packed in beside the headline. Farther down, the colored labels in the insurance section had awkward spacing. If I&amp;apos;m paying for a Max plan, those are things I&amp;apos;d expect it to get right without having to go back and ask for fixes.&lt;/p&gt;&lt;p&gt;Astra gave him a brand concept with a separate guide and a homepage. The colors felt more considered to me, and the orange it chose for actions actually carried through to the buttons. It also used image generation to fill in the page.&lt;/p&gt;&lt;p&gt;What stood out to me was how much Astra filled in on its own, including using the image generator without Ryan having to ask for it separately. Ryan has built a site he likes with Claude too, but in this comparison, Astra gave him more of what he&amp;apos;d need to keep building the business.&lt;/p&gt;&lt;h2&gt;I care what it costs to finish the job&lt;/h2&gt;&lt;p&gt;I&amp;apos;m not paying for a direct Anthropic subscription. I access its models through tools such as Cursor and OpenRouter. At the time we recorded, I was still on my existing $200 ChatGPT Pro plan. Ryan was comparing his $100 ChatGPT Pro and $100 Claude Max subscriptions.&lt;/p&gt;&lt;p&gt;One correction to our conversation: standard API pricing is the same for these two models, $10 per million input tokens and $50 per million output tokens. That is what OpenAI lists for Astra and Anthropic lists for Fable 5.1 . App subscriptions and third-party credits are different billing systems.&lt;/p&gt;&lt;p&gt;The total cost of a task still depends on how much the model uses to finish it. A higher price per token doesn&amp;apos;t automatically mean a more expensive finished page, and the same token price doesn&amp;apos;t guarantee the same bill.&lt;/p&gt;&lt;p&gt;The subscription limits also matter to me. Claude&amp;apos;s pricing page lists Fable usage at 50% of weekly limits on Max. That specifically limits how much of the weekly allowance can go to Fable. OpenAI says new $200 Pro sign-ups are temporarily paused , while existing subscriptions continue. These are the plan details I checked for September 15, 2026.&lt;/p&gt;&lt;p&gt;I also overstated it when I talked about having no limit on Astra. Astra has usage limits , and the app and plan matter. My preference is about the work I can get through with my subscription.&lt;/p&gt;&lt;h2&gt;The best model isn&amp;apos;t doing all my work&lt;/h2&gt;&lt;p&gt;I use GPT-5.6 Luna as much as any model. If the task is moving information around or doing straightforward work, I don&amp;apos;t need Fable or Astra for every step.&lt;/p&gt;&lt;p&gt;Ryan uses GPT-5.6 Sol to check work produced by other models. For a high-value task, he may use Astra for that final check. My addition was that the checking model needs the skills and context to know what we care about. Otherwise, we&amp;apos;re explaining ourselves again every time.&lt;/p&gt;&lt;p&gt;That&amp;apos;s the same reason I keep my instructions outside a single app, which I wrote about in my Grok Bot review . I want to carry the way I work with me when I switch tools.&lt;/p&gt;&lt;h2&gt;A privacy correction worth making&lt;/h2&gt;&lt;p&gt;I was too broad in the recording when I talked about Anthropic keeping data and turning training off in ChatGPT. Training and retention are separate questions.&lt;/p&gt;&lt;p&gt;ChatGPT lets consumer users opt out of training , and Anthropic gives consumer users a training choice . Their API and business products generally don&amp;apos;t train on customer content by default. That doesn&amp;apos;t mean nothing is retained. Anthropic has specific retention requirements for covered models such as Fable 5.1 . OpenAI also describes retention separately from training in its API data controls . The rules of the app or intermediary I&amp;apos;m using matter too.&lt;/p&gt;&lt;p&gt;I still care where my private thoughts and ideas go. I just shouldn&amp;apos;t treat a training toggle as a promise that nothing is saved.&lt;/p&gt;&lt;h2&gt;Is GPT-6 Astra or Claude Fable 5.1 better for writing?&lt;/h2&gt;&lt;p&gt;I preferred Fable 5.1&amp;apos;s hook and structure when I gave both models 17 scanned pages of my handwritten ideas. Astra still produced good writing and finished the whole job much faster, so I&amp;apos;d use Fable when that extra help with the writing is worth the wait.&lt;/p&gt;&lt;h2&gt;Which model did better on the website tests?&lt;/h2&gt;&lt;p&gt;In Ryan&amp;apos;s tests, Fable stayed closer to his existing landing page&amp;apos;s branding, while Astra made a shorter page that I thought was worth testing with customers. For the basement-cleanup business, Astra gave him a brand concept and guide along with the homepage and a generated image. I preferred that result, but we hadn&amp;apos;t measured which page converted better.&lt;/p&gt;&lt;h2&gt;Should I choose ChatGPT Pro or Claude Max?&lt;/h2&gt;&lt;p&gt;If I&amp;apos;m recommending one subscription, I&amp;apos;d choose ChatGPT Pro. Fable&amp;apos;s writing can be better, but I get more work done for my money with OpenAI&amp;apos;s models and tools. That makes ChatGPT Pro the better value for how I work.&lt;/p&gt;&lt;h2&gt;Do Astra and Fable have different API prices?&lt;/h2&gt;&lt;p&gt;At the September 15 comparison, both listed standard API rates of $10 per million input tokens and $50 per million output tokens. The cost of finishing a task can still differ because the models may use different amounts of tokens. Those rates are separate from a ChatGPT Pro or Claude Max subscription.&lt;/p&gt;&lt;h2&gt;Do I need Astra or Fable for every task?&lt;/h2&gt;&lt;p&gt;I use GPT-5.6 Luna as much as any model because straightforward work doesn&amp;apos;t always need the most expensive option. Ryan uses GPT-5.6 Sol to check other models&amp;apos; work and may use Astra for a higher-value task. I want the model doing that check to already have my skills and context.&lt;/p&gt;&lt;h2&gt;OpenAI is doing a better job of marketing this&lt;/h2&gt;&lt;p&gt;I also think OpenAI is doing a better job of winning over the average person with its marketing on Twitter, LinkedIn, and Instagram, along with things like the box of Astra hats they sent me. Most people aren&amp;apos;t comparing models the way Ryan and I are; they know which commercial was cooler or which ad got them to click. Right now, it feels like OpenAI has the momentum there too.&lt;/p&gt;&lt;p&gt;Ultimately, OpenAI&amp;apos;s subscriptions are a better value for the work I need to get done, even though Fable is better at some of it. If someone is only choosing one subscription, I&amp;apos;d tell them to choose ChatGPT Pro over Claude Max because they can get more done for their money.&lt;/p&gt;&lt;h2&gt;Questions&lt;/h2&gt;&lt;h3&gt;Is GPT-6 Astra or Claude Fable 5.1 better for writing?&lt;/h3&gt;&lt;p&gt;I preferred Fable 5.1&amp;apos;s hook and structure when I gave both models 17 scanned pages of my handwritten ideas. Astra still produced good writing and finished the whole job much faster, so I&amp;apos;d use Fable when that extra help with the writing is worth the wait.&lt;/p&gt;&lt;h3&gt;Which model did better on the website tests?&lt;/h3&gt;&lt;p&gt;In Ryan&amp;apos;s tests, Fable stayed closer to his existing landing page&amp;apos;s branding, while Astra made a shorter page that I thought was worth testing with customers. For the basement-cleanup business, Astra gave him a brand concept and guide along with the homepage and a generated image. I preferred that result, but we hadn&amp;apos;t measured which page converted better.&lt;/p&gt;&lt;h3&gt;Should I choose ChatGPT Pro or Claude Max?&lt;/h3&gt;&lt;p&gt;If I&amp;apos;m recommending one subscription, I&amp;apos;d choose ChatGPT Pro. Fable&amp;apos;s writing can be better, but I get more work done for my money with OpenAI&amp;apos;s models and tools. That makes ChatGPT Pro the better value for how I work.&lt;/p&gt;&lt;h3&gt;Do Astra and Fable have different API prices?&lt;/h3&gt;&lt;p&gt;At the September 15 comparison, both listed standard API rates of $10 per million input tokens and $50 per million output tokens. The cost of finishing a task can still differ because the models may use different amounts of tokens. Those rates are separate from a ChatGPT Pro or Claude Max subscription.&lt;/p&gt;&lt;h3&gt;Do I need Astra or Fable for every task?&lt;/h3&gt;&lt;p&gt;I use GPT-5.6 Luna as much as any model because straightforward work doesn&amp;apos;t always need the most expensive option. Ryan uses GPT-5.6 Sol to check other models&amp;apos; work and may use Astra for a higher-value task. I want the model doing that check to already have my skills and context.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>I paid $300 for Grok Bot. Its cloud computer is the most useful part.</title>
    <link>https://aaronmakelky.com/blog/grok-bot-worth-it-300-dollar-review</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/grok-bot-worth-it-300-dollar-review</guid>
    <pubDate>Tue, 08 Sep 2026 12:00:00 GMT</pubDate>
    <description>I use Grok Bot every day. Here&apos;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.</description>
    <content:encoded>&lt;h2&gt;Article&lt;/h2&gt;&lt;p&gt;From my appearance on AI Rabbit Holes with Ryan Doser.&lt;/p&gt;&lt;p&gt;I paid full price for Grok Bot&amp;apos;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.&lt;/p&gt;&lt;p&gt;When Ryan Doser asked whether I would recommend buying the subscription just for Grok Bot, my answer was no. I&amp;apos;m still hitting usage limits on the highest plan.&lt;/p&gt;&lt;p&gt;I talked with Ryan on AI Rabbit Holes about Grok Bot, the Slack-style agent product from xAI. We covered setup, rate limits, who it&amp;apos;s for, and the other tools included in the subscription.&lt;/p&gt;&lt;h2&gt;I can organize bots in channels, but I can&amp;apos;t invite a colleague&lt;/h2&gt;&lt;p&gt;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&amp;apos;t bring a human colleague into the workspace to work alongside me.&lt;/p&gt;&lt;p&gt;I find the setup much simpler than OpenClaw. Grok Bot runs in the cloud, so I don&amp;apos;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&amp;apos;t stop the work.&lt;/p&gt;&lt;p&gt;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.&lt;/p&gt;&lt;p&gt;For someone confused by the different chat, desktop, and coding apps, I can see the appeal of accessing their connected tools through one app.&lt;/p&gt;&lt;h2&gt;I still hit the limits on the highest plan&lt;/h2&gt;&lt;p&gt;The entry price keeps coming down, but the limits on the cheaper plans, including free access, concern me. I don&amp;apos;t think access is especially useful if someone can run out of usage before finishing the setup they need.&lt;/p&gt;&lt;p&gt;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.&lt;/p&gt;&lt;p&gt;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.&lt;/p&gt;&lt;p&gt;I&amp;apos;m on the highest plan and still hitting weekly caps. Based on that experience, I don&amp;apos;t recommend expecting a full week of heavy work across multiple projects. Moderate personal use seems more realistic to me.&lt;/p&gt;&lt;h2&gt;I can&amp;apos;t choose the model or its reasoning level&lt;/h2&gt;&lt;p&gt;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&amp;apos;t give me those controls. Its interface also doesn&amp;apos;t tell me the reasoning level it&amp;apos;s using.&lt;/p&gt;&lt;p&gt;I understand why someone else might prefer automatic model selection. If they don&amp;apos;t follow model releases, they may want to make a request and let the app choose. I find Notion&amp;apos;s model menu overwhelming enough that I can see why someone would leave that decision alone.&lt;/p&gt;&lt;p&gt;My frustration is that I can&amp;apos;t assign easier work to a different model, even within the company&amp;apos;s own models. I want to reserve the more capable model for tasks that need it and have more control over my usage.&lt;/p&gt;&lt;p&gt;That limitation makes Grok Bot a poor choice as my main AI tool. Someone who prefers automatic selection may care less about it.&lt;/p&gt;&lt;h2&gt;My bots have specific responsibilities&lt;/h2&gt;&lt;p&gt;I don&amp;apos;t like calling a setup &amp;quot;best practices&amp;quot; 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.&lt;/p&gt;&lt;p&gt;Every morning while I&amp;apos;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&amp;apos;t have to send a message to start it.&lt;/p&gt;&lt;p&gt;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.&lt;/p&gt;&lt;p&gt;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&amp;apos;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.&lt;/p&gt;&lt;p&gt;My bots can prepare drafts in Gmail, but I don&amp;apos;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.&lt;/p&gt;&lt;h2&gt;I review the podcast edit before publishing&lt;/h2&gt;&lt;p&gt;I showed Ryan a podcast example using connectors for the tools where I edit and publish.&lt;/p&gt;&lt;p&gt;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&amp;apos;t perfect.&lt;/p&gt;&lt;p&gt;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.&lt;/p&gt;&lt;p&gt;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.&lt;/p&gt;&lt;h2&gt;I can use the cloud computer myself&lt;/h2&gt;&lt;p&gt;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&amp;apos;s command-line tool in its terminal.&lt;/p&gt;&lt;p&gt;The power went out at my house the night before we recorded. Work running on the remote computer can continue independently of my Mac.&lt;/p&gt;&lt;p&gt;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.&lt;/p&gt;&lt;p&gt;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&amp;apos;t consume Grok Bot&amp;apos;s allowance as quickly.&lt;/p&gt;&lt;p&gt;Using only the bots in the chat leaves part of my subscription unused. I find Cursor&amp;apos;s cloud agents and Grok Build more useful for some of that work.&lt;/p&gt;&lt;h2&gt;I wouldn&amp;apos;t switch from a setup that already works&lt;/h2&gt;&lt;p&gt;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.&lt;/p&gt;&lt;p&gt;Someone who wants to customize and troubleshoot an agent system may prefer OpenClaw. I&amp;apos;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.&lt;/p&gt;&lt;p&gt;I haven&amp;apos;t seen business teams replacing Slack with Grok Bot. It doesn&amp;apos;t support bringing human colleagues into the workspace, and its usage limits make me doubtful about relying on it for heavy business use.&lt;/p&gt;&lt;p&gt;If Claude Code and Codex are already working well for someone, I don&amp;apos;t see a strong reason to switch. A simpler interface for accessing work while away from the computer might help, but I wouldn&amp;apos;t rebuild an effective setup just because Grok Bot is getting attention.&lt;/p&gt;&lt;p&gt;On X , 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&amp;apos;t answer those questions, and those capabilities are already available through other tools.&lt;/p&gt;&lt;p&gt;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&amp;apos;t want the instructions for my work tied to whichever app has a feature first.&lt;/p&gt;&lt;p&gt;I also think Cursor&amp;apos;s marketing does a poor job of explaining its uses to people who aren&amp;apos;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.&lt;/p&gt;&lt;h2&gt;Putting these tools to work with your team&lt;/h2&gt;&lt;p&gt;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.&lt;/p&gt;&lt;p&gt;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.&lt;/p&gt;&lt;p&gt;To get help bringing tools like this into your business or team, you can book a 30-minute call with me . We can talk about what you&amp;apos;re trying to get done and where it makes sense to start.&lt;/p&gt;&lt;p&gt;Where to find Ryan Doser:&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>I&apos;m hosting.</title>
    <link>https://aaronmakelky.com/blog/im-hosting</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/im-hosting</guid>
    <pubDate>Fri, 04 Sep 2026 12:00:00 GMT</pubDate>
    <description>Astra Commons is coming to Casper. Join me and OpenAIDevs on September 12 at CitiWerks for free coffee, food, and GPT-6 credits.</description>
    <content:encoded>&lt;h2&gt;Astra Commons is coming to Casper&lt;/h2&gt;&lt;p&gt;Aaron Makelky is hosting Astra Commons in Casper, Wyoming with OpenAIDevs. Saturday, September 12, 10:00 a.m. to noon at CitiWerks Cafe and Bistro.&lt;/p&gt;&lt;p&gt;It&amp;apos;s a casual meetup for developers, students, founders, and community members around the GPT-6 Astra launch. Laptops are optional. Space is limited.&lt;/p&gt;&lt;h2&gt;What registered attendees get&lt;/h2&gt;&lt;p&gt;Free coffee and food, time with the local builder community, and $100 in Codex credits plus $50 in API credits to build with GPT-6 Astra.&lt;/p&gt;&lt;p&gt;Register on Luma. The venue is CitiWerks Cafe and Bistro at 162 S Wolcott St, Casper.&lt;/p&gt;&lt;h2&gt;Questions&lt;/h2&gt;&lt;h3&gt;When is Astra Commons Casper?&lt;/h3&gt;&lt;p&gt;Saturday, September 12, 2026, from 10:00 a.m. to noon Mountain Time.&lt;/p&gt;&lt;h3&gt;Where is Astra Commons Casper?&lt;/h3&gt;&lt;p&gt;CitiWerks Cafe and Bistro, 162 S Wolcott St, Casper, Wyoming.&lt;/p&gt;&lt;h3&gt;Is Astra Commons Casper free?&lt;/h3&gt;&lt;p&gt;Yes. Registration is free. Registered attendees also get Codex and API credits.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>How to get more Grok Bot usage (without paying more)</title>
    <link>https://aaronmakelky.com/blog/how-to-get-more-grok-bot-usage</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/how-to-get-more-grok-bot-usage</guid>
    <pubDate>Thu, 03 Sep 2026 12:00:00 GMT</pubDate>
    <description>Grok Bot is awesome, but the rate limits suck. Here&apos;s a little trick that&apos;s totally legit: run Cursor, Grok Build, Codex, or Google&apos;s anti-gravity on Grok Bot&apos;s own Linux terminal so the coding work does not eat your usage.</description>
    <content:encoded>&lt;h2&gt;Grok Bot runs on a virtual server&lt;/h2&gt;&lt;p&gt;Grok Bot is awesome, but the rate limits suck. If Grok Bot itself handles all the jobs you give it, you burn through usage fast. Two days into the week I was already at 64%.&lt;/p&gt;&lt;p&gt;Grok Bot runs on a Linux device in the cloud with a terminal. When you pass it a coding task, that work uses your Grok Bot rate limits.&lt;/p&gt;&lt;h2&gt;You can install coding agents&lt;/h2&gt;&lt;p&gt;On that terminal you can install coding agents. The same Grok 4.6 model can run in Cursor on the command line and take from your Cursor rate limits, not Grok Bot. Same with Grok Build, Codex, or Google&amp;apos;s anti-gravity.&lt;/p&gt;&lt;p&gt;You do not have to use all of them. For any ones you have a sub for, tell Grok Bot to open the terminal, use that tool, do the coding task, and push the work to GitHub.&lt;/p&gt;&lt;h2&gt;Give Herdr to your agent&lt;/h2&gt;&lt;p&gt;Herdr manages multiple terminal sessions. You can give that agent management tool to Grok Bot so it herds Codex, Grok, and anti-gravity on its own virtual machine and reports back.&lt;/p&gt;&lt;p&gt;The only rate limits that come out of your Grok Bot usage are from the prompting it is doing, not the actual coding task.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>Your Grok Bot setup is a mess</title>
    <link>https://aaronmakelky.com/blog/how-i-set-up-grok-bot</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/how-i-set-up-grok-bot</guid>
    <pubDate>Mon, 31 Aug 2026 12:00:00 GMT</pubDate>
    <description>I organize Grok Bot like an org system with a chief of staff, project leads, and shared specialists.</description>
    <content:encoded>&lt;h2&gt;The org system&lt;/h2&gt;&lt;p&gt;I organize Grok Bot like an org system with a chief of staff, project leads, and shared specialists.&lt;/p&gt;&lt;p&gt;The chief is the main interface. Project leaders own projects, not tasks. Specialists sit underneath those leads and get reused across projects.&lt;/p&gt;&lt;p&gt;If you add a bunch of bots on one flat level, it does not work. Pin the chief and active project leads. Hide the one-off bots.&lt;/p&gt;&lt;h2&gt;Colors, shapes, and channels&lt;/h2&gt;&lt;p&gt;Use the default Grok Bot designs. My chief is the only purple bot. Project leaders each get a unique color. Specialists are green.&lt;/p&gt;&lt;p&gt;Cloud for the chief, triangles for leads, squares for specialists. I do not use the other shapes.&lt;/p&gt;&lt;p&gt;You need two kinds of channels: all hands, where the chief coordinates the project managers every weekday, and one project channel per triangle bot.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>I edited a fishing video from the ChatGPT mobile app without touching a timeline</title>
    <link>https://aaronmakelky.com/blog/codex-hyperframes-video-editing-workflow</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/codex-hyperframes-video-editing-workflow</guid>
    <pubDate>Sat, 25 Jul 2026 12:00:00 GMT</pubDate>
    <description>The real prompts, screenshots, corrections, and tools behind a 31.7-second vertical video I edited with Codex and HyperFrames from my phone.</description>
    <content:encoded>&lt;h2&gt;How the video was made&lt;/h2&gt;&lt;p&gt;Aaron gave Codex one Google Drive video, ten fishing photos, a screenshot of a Codex fishing report, and access to his existing video brand system.&lt;/p&gt;&lt;p&gt;Codex transcribed the source, selected three spoken sections, assembled a five-scene 1080 by 1920 HyperFrames project, and returned review MP4s to the ChatGPT mobile app.&lt;/p&gt;&lt;p&gt;Aaron reviewed each cut on his iPhone and sent specific visual corrections until the final 31.7-second edit was ready for X, Instagram, and TikTok.&lt;/p&gt;&lt;h2&gt;Tools in the workflow&lt;/h2&gt;&lt;p&gt;The workflow used ChatGPT mobile remote, OpenAI Codex, Google Drive, GitHub, HyperFrames 0.7.71, FFmpeg, FFprobe, OCR, MiniMax CLI, and a licensed Melodie music track.&lt;/p&gt;&lt;p&gt;MiniMax music generation failed because the connected account had no active token plan, so the final edit used music Aaron had already licensed through Melodie.&lt;/p&gt;&lt;h2&gt;Questions&lt;/h2&gt;&lt;h3&gt;Did Aaron edit the video in a traditional video editor?&lt;/h3&gt;&lt;p&gt;No. Aaron reviewed the video and sent corrections through the ChatGPT mobile app. Codex changed and rendered the HyperFrames project on his Mac.&lt;/p&gt;&lt;h3&gt;What did HyperFrames do in this workflow?&lt;/h3&gt;&lt;p&gt;HyperFrames handled the code-based timeline, scene composition, on-screen graphics, captions, photo timing, transitions, checks, and final MP4 render.&lt;/p&gt;&lt;h3&gt;Can this workflow create a version for native social captions?&lt;/h3&gt;&lt;p&gt;Yes. The project includes a native-social export that removes the music and narration captions while preserving Aaron&amp;apos;s voice, river audio, story graphics, and timing.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>one card run beat a year of errands</title>
    <link>https://aaronmakelky.com/what-i-built-today-2026-06-16</link>
    <guid isPermaLink="true">https://aaronmakelky.com/what-i-built-today-2026-06-16</guid>
    <pubDate>Tue, 16 Jun 2026 12:00:00 GMT</pubDate>
    <description>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.</description>
    <content:encoded>&lt;h2&gt;Build log summary&lt;/h2&gt;&lt;p&gt;A night-before anniversary card run showed the timing problem. The calendar had the dates, but the card errand was still happening too late. The fix was a Poke recipe that gathers card-worthy dates into one chronological list, then one dollar-store trip to buy the cards in a batch.&lt;/p&gt;&lt;p&gt;The useful pattern was turning the calendar into a card-shopping list before the errand became another night-before scramble.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>the row shape was the build</title>
    <link>https://aaronmakelky.com/what-i-built-today-2026-06-15</link>
    <guid isPermaLink="true">https://aaronmakelky.com/what-i-built-today-2026-06-15</guid>
    <pubDate>Mon, 15 Jun 2026 12:00:00 GMT</pubDate>
    <description>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.</description>
    <content:encoded>&lt;h2&gt;Build log summary&lt;/h2&gt;&lt;p&gt;A multi-channel promo handoff was starting to create extra work because each platform had its own row. The cleanup made the publishing moment the durable unit: one page for the slot, with channel versions and creative kept together.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>the chart needed a memory</title>
    <link>https://aaronmakelky.com/what-i-built-today-2026-06-04</link>
    <guid isPermaLink="true">https://aaronmakelky.com/what-i-built-today-2026-06-04</guid>
    <pubDate>Thu, 04 Jun 2026 12:00:00 GMT</pubDate>
    <description>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.</description>
    <content:encoded>&lt;h2&gt;Build log summary&lt;/h2&gt;&lt;p&gt;The visible task was one monthly slide update. The useful build was the packet around it: the chart, the numbers, the short read, and the final manual paste step all in one place.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>the comment had no instructions</title>
    <link>https://aaronmakelky.com/what-i-built-today-2026-06-02</link>
    <guid isPermaLink="true">https://aaronmakelky.com/what-i-built-today-2026-06-02</guid>
    <pubDate>Tue, 02 Jun 2026 12:00:00 GMT</pubDate>
    <description>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.</description>
    <content:encoded>&lt;h2&gt;Build log summary&lt;/h2&gt;&lt;p&gt;A vague assigned comment in a recurring review deck turned into a concrete update packet. The agent found the real ask, pulled analytics, rendered the chart, drafted the slide copy, and stopped at the manual paste-and-resolve step instead of pretending the whole job was done.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>I did not have to ask</title>
    <link>https://aaronmakelky.com/what-i-built-today-2026-05-29</link>
    <guid isPermaLink="true">https://aaronmakelky.com/what-i-built-today-2026-05-29</guid>
    <pubDate>Fri, 29 May 2026 12:00:00 GMT</pubDate>
    <description>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.</description>
    <content:encoded>&lt;h2&gt;Build log summary&lt;/h2&gt;&lt;p&gt;Today worked because the board had already sorted the day before I touched it. The useful build was a dashboard that showed what could move now and what had to wait, so I stopped wasting time on the rest.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>no specialists in the room</title>
    <link>https://aaronmakelky.com/what-i-built-today-2026-05-28</link>
    <guid isPermaLink="true">https://aaronmakelky.com/what-i-built-today-2026-05-28</guid>
    <pubDate>Thu, 28 May 2026 12:00:00 GMT</pubDate>
    <description>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.</description>
    <content:encoded>&lt;h2&gt;Build log summary&lt;/h2&gt;&lt;p&gt;Three specialist jobs landed in one day. The motion designer was on PTO, so I built the piece in Jitter from a template and Descript posted it the same day. A livestream viewer asked what a PM said about a specific set of features, so I pulled the transcript through the Descript API, found the moment, and DM&amp;apos;d the answer back on LinkedIn. A workshop&amp;apos;s follow-up needed custom replies based on each attendee&amp;apos;s session feedback, so I ran the entire loop inside Notion: form, table, drafted replies, Notion Mail and Gmail drafts. No external SaaS.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>the dashboard asked first</title>
    <link>https://aaronmakelky.com/what-i-built-today-2026-05-27</link>
    <guid isPermaLink="true">https://aaronmakelky.com/what-i-built-today-2026-05-27</guid>
    <pubDate>Wed, 27 May 2026 12:00:00 GMT</pubDate>
    <description>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.</description>
    <content:encoded>&lt;h2&gt;Build log summary&lt;/h2&gt;&lt;p&gt;Today worked because the dashboard had already surfaced the few projects that mattered, including a same-day workshop lane and a next-day event lane. The useful build was the operating card that let me shape the work instead of hunting for it.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>the dashboard got there first</title>
    <link>https://aaronmakelky.com/what-i-built-today-2026-05-26</link>
    <guid isPermaLink="true">https://aaronmakelky.com/what-i-built-today-2026-05-26</guid>
    <pubDate>Tue, 26 May 2026 12:00:00 GMT</pubDate>
    <description>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.</description>
    <content:encoded>&lt;h2&gt;Build log summary&lt;/h2&gt;&lt;p&gt;Today worked because the real priorities were already waiting when I opened the dashboard. From there I tightened one workshop into a single board, cut a content plan down to the useful version, and moved the operating layer into a repo agents could read so the same system can travel across tools and devices.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>the brief was the build</title>
    <link>https://aaronmakelky.com/what-i-built-today-2026-05-25</link>
    <guid isPermaLink="true">https://aaronmakelky.com/what-i-built-today-2026-05-25</guid>
    <pubDate>Mon, 25 May 2026 12:00:00 GMT</pubDate>
    <description>A public build log about using AI to turn a fuzzy workshop session into a useful brief and a smaller demo.</description>
    <content:encoded>&lt;h2&gt;Build log summary&lt;/h2&gt;&lt;p&gt;Sunday night I did the least glamorous version of AI work. I took an internal session that was still a little foggy and turned it into a page I could actually drive Monday from.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>four useful AI reps, one of them about a hoodie</title>
    <link>https://aaronmakelky.com/what-i-built-today-2026-05-22</link>
    <guid isPermaLink="true">https://aaronmakelky.com/what-i-built-today-2026-05-22</guid>
    <pubDate>Fri, 22 May 2026 12:00:00 GMT</pubDate>
    <description>A public build log about practical AI watcher work, connector distribution, agent planning, and context carry-forward.</description>
    <content:encoded>&lt;h2&gt;Build log summary&lt;/h2&gt;&lt;p&gt;A couple of these started yesterday. I am still counting them because they were real work that shipped, and they tell a better story about how I actually use AI at work than a polished status report could.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>I cleaned up ten skill folders so you can see the wiring</title>
    <link>https://aaronmakelky.com/blog/10-openclaw-skills-running-on-raspberry-pi</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/10-openclaw-skills-running-on-raspberry-pi</guid>
    <pubDate>Tue, 07 Apr 2026 12:00:00 GMT</pubDate>
    <description>Hardware is a Pi 5. Each folder has a SKILL.md, a SETUP.md, and scripts only where a prompt is not enough.</description>
    <content:encoded>&lt;h2&gt;Article summary&lt;/h2&gt;&lt;p&gt;This post walks through the OpenClaw skill folders from my Raspberry Pi setup. The useful part is the pattern: small folders with setup docs, scripts where the job needs them, schedules, and notes about where things break.&lt;/p&gt;&lt;p&gt;The bundle includes Morning Briefing, Daily Video Learning, Weekly Review, Nightly YOLO, Content Flywheel, Geopolitical Briefing, Social Media Intake, Weather, Recipe Processor, and Google Calendar.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>He told me he felt like he could finally read</title>
    <link>https://aaronmakelky.com/blog/ai-in-education-friction-question-not-tool-question</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/ai-in-education-friction-question-not-tool-question</guid>
    <pubDate>Fri, 27 Mar 2026 12:00:00 GMT</pubDate>
    <description>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.</description>
    <content:encoded>&lt;h2&gt;Article summary&lt;/h2&gt;&lt;p&gt;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.&lt;/p&gt;&lt;p&gt;He told me he felt like he could finally read is grouped under Classroom AI, Literacy, Podcast.&lt;/p&gt;&lt;p&gt;Use the post as a starting point, then test one small workflow before rebuilding a whole system.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>revision history may not catch the agent</title>
    <link>https://aaronmakelky.com/blog/agentic-ai-classrooms-what-teachers-need-to-know</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/agentic-ai-classrooms-what-teachers-need-to-know</guid>
    <pubDate>Fri, 20 Feb 2026 12:00:00 GMT</pubDate>
    <description>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.</description>
    <content:encoded>&lt;h2&gt;Article summary&lt;/h2&gt;&lt;p&gt;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.&lt;/p&gt;&lt;p&gt;revision history may not catch the agent is grouped under Classroom AI, Agentic AI.&lt;/p&gt;&lt;p&gt;Use the post as a starting point, then test one small workflow before rebuilding a whole system.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>cheaper tools did not make anyone sound like themselves</title>
    <link>https://aaronmakelky.com/blog/podcasting-for-creators-human-content-wins-2026</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/podcasting-for-creators-human-content-wins-2026</guid>
    <pubDate>Fri, 20 Feb 2026 12:00:00 GMT</pubDate>
    <description>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.</description>
    <content:encoded>&lt;h2&gt;Article summary&lt;/h2&gt;&lt;p&gt;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.&lt;/p&gt;&lt;p&gt;cheaper tools did not make anyone sound like themselves is grouped under Podcasting, Creators.&lt;/p&gt;&lt;p&gt;Use the post as a starting point, then test one small workflow before rebuilding a whole system.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>most teacher AI apps are wrappers</title>
    <link>https://aaronmakelky.com/blog/3_ai_apps_teachers</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/3_ai_apps_teachers</guid>
    <pubDate>Mon, 03 Mar 2025 12:00:00 GMT</pubDate>
    <description>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.</description>
    <content:encoded>&lt;h2&gt;Article summary&lt;/h2&gt;&lt;p&gt;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.&lt;/p&gt;&lt;p&gt;most teacher AI apps are wrappers is grouped under Classroom AI, Tools.&lt;/p&gt;&lt;p&gt;Use the post as a starting point, then test one small workflow before rebuilding a whole system.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>context, a revision, then check the work</title>
    <link>https://aaronmakelky.com/blog/ai_skills_everyone_2025</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/ai_skills_everyone_2025</guid>
    <pubDate>Sat, 22 Feb 2025 12:00:00 GMT</pubDate>
    <description>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.</description>
    <content:encoded>&lt;h2&gt;Article summary&lt;/h2&gt;&lt;p&gt;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.&lt;/p&gt;&lt;p&gt;context, a revision, then check the work is grouped under Classroom AI, Habits.&lt;/p&gt;&lt;p&gt;Use the post as a starting point, then test one small workflow before rebuilding a whole system.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>AI policy for schools without pretending it is simple</title>
    <link>https://aaronmakelky.com/blog/ai_policy_for_schools</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/ai_policy_for_schools</guid>
    <pubDate>Mon, 17 Feb 2025 12:00:00 GMT</pubDate>
    <description>A school AI policy has to answer normal classroom questions about what students can use and disclose, and where teacher judgment still sits.</description>
    <content:encoded>&lt;h2&gt;Article summary&lt;/h2&gt;&lt;p&gt;A school AI policy has to answer normal classroom questions about what students can use and disclose, and where teacher judgment still sits.&lt;/p&gt;&lt;p&gt;AI policy for schools without pretending it is simple is grouped under AI Policy, School Administration, Teaching.&lt;/p&gt;&lt;p&gt;Use the post as a starting point, then test one small workflow before rebuilding a whole system.&lt;/p&gt;</content:encoded>
  </item>
  <item>
    <title>I made these AI mistakes as a teacher first</title>
    <link>https://aaronmakelky.com/blog/5-mistakes-teachers-make-with-ai</link>
    <guid isPermaLink="true">https://aaronmakelky.com/blog/5-mistakes-teachers-make-with-ai</guid>
    <pubDate>Sun, 02 Feb 2025 12:00:00 GMT</pubDate>
    <description>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.</description>
    <content:encoded>&lt;h2&gt;Article summary&lt;/h2&gt;&lt;p&gt;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.&lt;/p&gt;&lt;p&gt;I made these AI mistakes as a teacher first is grouped under Classroom AI, Teaching.&lt;/p&gt;&lt;p&gt;Use the post as a starting point, then test one small workflow before rebuilding a whole system.&lt;/p&gt;</content:encoded>
  </item>
  </channel>
</rss>
