Updated July 2026
What I'm doing now
A snapshot of what has my attention this quarter. It's a now page, so it goes stale on purpose. If the date above is old, so is the list.
Building
Small tools I actually use: an evaluation harness I keep reaching for before I trust a model with real work, and a set of internal helpers that keep agent runs inspectable: durable logs, small reversible steps, and a status line I can actually read instead of a chat transcript.
Testing
The AI coding tools, in anger, on real projects (Claude Code, Cursor, Copilot), keeping honest receipts on what each one costs against what it saves. Right now most of the interesting questions are about agents: where letting one write, refactor, and review genuinely helps, and where it quietly makes more work than it removes.
Chewing on
How to keep an AI-assisted codebase legible six months later: the evals, the guardrails, and the small habits that stop "the model wrote it" from becoming "nobody understands it." That's most of what ends up in the writing and the notes.