On AI product strategy, LLM engineering, and shipping agentic systems that survive production.
Loop engineering replaced prompt engineering as the core agent skill. The loop is twenty lines of code. The verifier inside it is the actual system.
Four things move underneath every AI feature, and only one has a version number. A field guide to drift, and why your eval set is the last thing to know.
Every competitor can buy the same model. None of them can buy an understanding of which constraint your organization was built around, or the speed to rebuild when it dissolves.
AI teams keep tuning the prompt when the real leverage is a feedback loop or the goal itself. Donella Meadows explained why decades ago. Here's the map.
A practical path from traditional PM to AI PM: what changes in your product instincts, the portfolio that gets you hired, realistic timelines, and current pay.
SDLC ships deterministic code. ADLC ships probabilistic behavior. That one shift rewrites what you build, how you test, and what done means.
No spam. Just the occasional note when I publish something worth your time.