I build software and make videos.

Founder of Little Worker · Austin, Texas

active today · 14,395 contributions/yr

I'm building Little Worker, software for working with AI agents.

I make videos about the things I build and the questions I keep coming back to. I keep asking until I understand, and share what I learn.

• Little-CISelf-hosted GitHub Actions runners in an isolated Ubuntu machine on Apple Silicon.2026↗
• Grok ResearchResearch and a local prototype exploring Grok Bot's desktop interface and agent execution.2026
• Nash AgentA TypeScript specification for using language models and game theory to reason through strategic decisions.2026↗
• Little WorkerSoftware for assigning work to AI agents, reviewing results, and keeping track of tasks and costs.2026
• Shorts FactoryQ4 2025 video-generation experiment built to learn Gemini, Veo, and automated production.2025

• You don't need to keep up with every AI toolSep 9, 2026
• 3B Tokens a Day: Why Workflows Beat PromptsApr 1, 2026 · daily
• Codex Makes Thumbnails and Graphics in Figma NowMar 24, 2026 · daily
• The New Programming Language is Markdown (prompts = subagents = skills)Mar 24, 2026 · daily

• This week in agents: 13 things that changedSep 22, 2026 · 7 min
• I use ChatGPT projects like small APIs for my repositoriesSep 13, 2026 · 3 min
• Little-CI: self-hosted runners on Apple SiliconSep 13, 2026 · 1 min
• How Grok Bot works: I took apart version 0.47Sep 11, 2026 · 15 min

• #3008fix(alerts): alert tasks have a 50 cent limit+63 −22h
• #2996fix(console): the free workspace archive no longer reports owners without email as failures+5 −52h
• #2994fix(release): the web deploys get the Cloudflare token from the release settings+12 −32h
• #2993fix(web): production source maps reach Sentry+11 −22h
• #2992fix(web): error reports name the browser again+20 −43h
• #2991fix(alerts): an alert task that is off or not made is logged, not reported to Sentry+57 −253h
• #3001fix(core): a hosted tools lookup that runs out of time is asked once more+66 −73h
• #3009docs: Managed Agents parity state, four answers, and a 128-item backlog+12,775 −513h
• #2995fix(infra): production keeps App Store purchases on in every deploy+26 −73h
• #2997chore(ios): the App Store page shows the meeting, giving a task, results and plans+345 −8473h
• #2998fix(core): the monitor writes a lasting problem at error priority only when it opens+43 −53h
• #3002test(console): the parallel allowance test asks again the call that stopped the last run+24 −63h
• #3000fix(core): a run held at a spend cap no longer skips the worker's daily routines+73 −14h
• #2999fix(core): a run held for a missing monthly limit says to set one+18 −114h
• #2989fix(console): the numbers page counts only real customers+96 −124h
• #2987fix(ios): the Subscription page no longer says the website costs less+34 −334h
• #2988docs(skills): one skill frees disk space, and a crit round deletes its screenshots+87 −05h
• #2986fix(core): web chats open again: browsers may send the client header+5 −05h
• #2984fix(core): the time charge sends one statement per turn again+40 −302d
• #2957fix(web): share cards show Pip on Sand, and the site card shows no Review+23 −142d