Writing

Essays & ideas

• 6 June 2026

  Fixing intent drift: a spec + governance layer for AI-assisted development

  AI agents write code brilliantly and forget why ruthlessly. Here's the system I built to make the why survive — and what happened when I ran a real production feature through it.

• 23 May 2026

  I didn't invent AppMaker. I curated it.

  None of the ideas in my governance layer are original — and that's the point. What I actually did was study five systems, take the single best idea from each, and say no to everything else. Here's the full provenance, including the two failures that taught me the most.

• 9 May 2026

  The agent is the new integration layer

  For decades the user was the real integration layer — copying values between windows, checking five systems to answer one question. A conversational agent with tools ends that: one interface, every application behind it. Anatomy of a real one, with 46 tools in production.

• 25 April 2026

  Building a grounded AI agent for a consultancy website

  A public-facing chat agent is a liability generator with a text box. Here's the architecture that keeps ours useful, honest and cheap — grounding, fail-closed budgets, and rate limits that get stricter when things break.

• 11 April 2026

  The four kinds of AI agents (and what's actually behind Claude Code)

  'Agent' has become a word for everything with a chat box. Useful taxonomy: what tools it holds, what it may write to, and how long it runs unsupervised. Four kinds, illustrated with systems I run in production — and a look inside a coding agent.

• 28 March 2026

  Talk to your AI agent like a caveman

  Politeness is for people. With agents, every word of padding competes with your actual instructions for the model's attention — and the research backs this up. Why terse beats verbose, and the 'caveman' practice that takes it to the logical extreme.