Half-formed thoughts on AI, agents, and the work that survives them — kept until they hold.

A verification gate is theater until someone can actually get fired for skipping it. Everything else — the checklist, the sign-off field, the "please review this" — is decoration around that one fact.

Cloudflare Wallets can link an agent payment to an account and constrain how it spends. A separate question remains: was an in-bounds purchase actually wanted?

A piece going around says the consultant's edge is moving from taste to the nerve to decide. Its "decisiveness" glues two things together, and only one of them is a skill AI can't already do.

"AI-native game" has one honest definition: pull the model out and see if play survives. Passing that test says nothing about whether the result has any taste.

I said taste was judgment renamed. Trying to pull the two words apart, I found I could — and the difference is what the moat argument keeps hiding.