Shownotes
This week I talk about AI experiments, and why most of them won't pay off. That's fine, as long as each one leaves something behind. I explain the test I'd use: is your tenth experiment cheaper and safer to run than your first? I walk through four pieces of shared infrastructure that make that true, what letting every team build on its own costs, how Goldman Sachs and Revolut handled it, and why breaking things still has value. Plus, Apple reportedly cutting its engineers' daily AI spending ...

