The pilot that never ends
A proof of concept has two honest endings: it ships or it gets killed. Renewal is neither, and it is the most common outcome I see.
Writing
Notes from inside the engagements.
Short pieces, published when I have something worth saying rather than on a schedule. No newsletter cadence, no growth funnel.
Write the evaluation before the prompt
If nobody on the team can say what a good answer looks like, model choice is not your problem yet. How to run the workshop that settles it.
What your data actually supports
Three questions that predict, better than anything else I've tried, whether a document-retrieval project will work.
Shadow usage is a signal, not a violation
People using assistants without permission are telling you exactly where the friction is. A ban throws that information away.
Against the AI strategy document
Strategy decks age badly at this pace. A short list of things you are building this quarter ages better and is harder to argue with.
Something here match a problem you have?
Reply to it in an email. I answer questions about these more reliably than I publish new ones.