03Think
Writing
Field notes and dated arguments. Short sentences. No adjectives about the author.
September 2, 2026
Why most AI testing pilots fail
They are scored on generation, starved of context, and shipped without a contract. The model is not the failure mode. The design of the pilot is.
4 min
August 18, 2026
What I learned running Claude inside a real QA pipeline
Selectors rot, assertions are the meaning, and the evaluator needs tests of its own. Notes from putting a model on work that ships.
4 min
July 9, 2026
The QA org chart of 2030
Fewer people writing tests. More people writing contracts. Agents as staff, with managers. A sketch of the function after generation is cheap.
3 min
June 4, 2026
Regulated industries will adopt AI testing first, not last
The usual story is that banks and hospitals move slowly. The usual story misses that they already have the one thing consumer tech keeps skipping: a way to answer an auditor.
3 min