Thinking
Short essays on AI, engineering and leadership, written from what I have actually shipped.
AI that an auditor can follow
In regulated engineering, the useful question is not whether AI is accurate. It is whether a person can see, check and own every decision it touches.
Before you rewrite it
A full rewrite feels like the brave choice. Usually it is the risky one. Notes on paying down technical debt while the product keeps shipping.
AI agents need a shared memory they are allowed to use
As engineering organizations adopt AI agents, each tool is quietly rebuilding its own copy of the same context. There is a better way.
Validation is the product
Most AI agent frameworks obsess over executing the next step. The ones that survive production obsess over when to stop.