On this page

Updated 19 November 2026

1 min read

Agentic AI

In my own time, on my own projects, I build software with a team of AI agents that I designed and run myself. It’s the most interesting thing I’ve learnt in years. It changes what one person can build.

How I work

  • Prototype first: a quick, throwaway version to find out what I actually want

  • Specification: an engineering agent turns the idea into a written specification and plan, which I review and approve before anything is built

  • Tests first: short-lived worker agents write the tests, then the code that passes them

  • Independent review: a fresh reviewer agent that never saw the build reads every change cold

  • Human-gated merge: I decide what gets built and I press the merge button myself

What I’ve done

  • Designed and run a self-hosted, multi-agent engineering team for my personal projects

  • Built this website and its Sphinx theme, sphinx-daisyui, with that team

  • Wrote the engineering standards the agents work to

Skills

  • Agentic software development and multi-agent workflows

  • Specification-driven and test-first development with AI

  • Review and quality gates for AI-written code

  • Running language models locally on my own hardware