The model is the most replaceable part of your stack
Whatever model is best today won't be in eighteen months. The teams that thrive treat that as an architectural requirement, not a nuisance — here's what that looks like in practice.
Writing
Notes from the field.
What we've learned building and fixing production AI systems, written down. No news, no hot takes — just the arguments we find ourselves making in engagements, in long form. We publish when we have something to say, which is not on a schedule.
If you can't measure it, you shipped a demo
The difference between an AI demo and an AI system is an evaluation harness. Why evals come first, what a useful one looks like, and the failure mode of skipping them.
Build, buy, or wait
Most AI build-versus-buy analyses compare the wrong things. A framework for the decision that includes the option vendors never mention: doing nothing yet.