Insights
Research and field notes on putting AI to work inside large organizations.
Research
What survived contact with real workloads: evaluation, tool design, and the true cost of a bad retry.
Read moreResearch
Role design, skills, and the operating model shifts that decide whether the tooling actually gets used.
Read moreResearch
Shipping the model is a quarter of the job. The rest is change management nobody budgeted for.
Read moreResearch
The line item is the small part. The real bill is the time your team spent not shipping, and the vendor you are now quietly locked into.
Read moreResearch
Teams postpone AI for a warehouse rebuild they do not need. What actually has to be true before an agent can do useful work, and what can wait.
Read moreResearch
How to evaluate an AI vendor when the pilot runs on their data and their happy path: evals on your own corpus, named failure modes, and behaviour at the tail.
Read more