Field notes on useful AI.

Most AI projects die at the demo. The ones that survive are narrower, quieter, and owned by the team that uses them.

What a good audit covers, what it costs, and the three questions that expose a bad one before you sign.

Prompt libraries rot. Systems that encode who approves what — those compound.

Dashboards answer questions nobody asked. Here is how we wire numbers to narratives.

Most agencies vanish after launch. A maintenance rhythm that keeps systems alive without a retainer trap.

A decision framework for the 80% of AI use-cases where the honest answer is 'not yet'.