Short essays, written to be argued with. Some are about the unglamorous parts of shipping AI inside an organization that has auditors — what "done" means, what makes autonomy sellable. The rest are about the other half of the job: how product teams get pointed in a direction, given the room to decide, and coached once they're there.
An AI feature without an eval set isn't untested — it's unspecified. Why the cases, the grading, and the bar belong in the spec, and what changes in a team when acceptance moves left.
Identity, auditability, and reversibility aren't a compliance pass at the end. They're the currency you buy agent autonomy with — and retrofitting them costs you a year of removing capability.
A deadlock between two competent teams usually means the problem has two regimes and nobody has named them. One question that finds the split instead of picking a winner.
What SVPG's product operating model and Teresa Torres's continuous discovery actually demand — and why installing the ceremonies without moving decision rights produces discovery theater.
Teams rarely lose to wrong decisions. They lose to undecided ones. An aspirational vision, a strategy tactical enough to rule things out, and naming which door you're standing in front of.
The weekly 1:1, the career conversation, and the performance conversation have different horizons and different owners. Running them as one meeting makes all three worse.
More in progress — on platform adoption, and on what actually changes when prototyping stops being expensive. If you want to argue with any of these, email me.