Writing
Notes on making AI work.
Practical pieces on building software end to end with AI, agents, retrieval, evaluation, and human review — plus a monthly reading of what the labs shipped. Opinionated, sourced, and written for people who build.
How I build software end to end with generative AI
AI agents write most of my code; I own the spec, the tests, the deploy, and the post-mortem. The six-step loop I use from idea to production, and the guardrails that keep it honest.
October 2026 in AI, week one: small models, cheaper caches, agents with an email address
Claude Haiku 5.5, halved cache-read prices, Mistral Large 4 in preview, Google's Gemini agent with its own workplace identity, and platforms tightening what agents may fetch.
September 2026 in AI: the month agents got their own computers
Frontier models moved down the price curve, always-on agents became products, and one lab published how much of its own research its model now leads. What changed, and what I'd do about it this quarter.
The evaluation gap: why agent demos don't survive production
A working demo tells you the happy path exists. An eval tells you how often you find it. How to build the golden set, grade traces, and gate every change.
Human-in-the-loop is a design decision, not a disclaimer
Three review patterns, a two-axis rule for choosing between them, and how to make the human step cheap enough that people actually do it.
Context engineering: a bigger window is not a strategy
What belongs in the context, what stays out, how to measure retrieval on your own questions, and what to preserve when you compact a long conversation.
MCP, plainly: what the Model Context Protocol actually standardizes
Hosts, servers, tools, and resources without the hype — plus the pitfalls (tool sprawl, injected results, over-broad scopes) that show up in production.
No posts under that topic yet.