Writing
How large companies actually get AI into production, what slows them down, and what to measure instead. New pieces arrive here first.
A role-level AI exposure score hides the decisions that matter. How we break each occupation into tasks, and what a paralegal's scores show.
Adapted from my LinkedIn posts of July 14, 15, 16 and 27, 2026
Four requirements and one immutable log decide whether enterprise AI survives the compliance review. They are a day-one architecture decision.
Adapted from my LinkedIn post of June 18, 2026
One number, for one pilot, with the waits written underneath it. How to measure the days between an accepted idea and the first live signal, and why you should never bonus it.
Adapted from Chapter 5 of The Seven Frictions
Does the vendor get paid when the workflow works, or when the engagement continues? The answer predicts how the project will go.
Adapted from my LinkedIn post of July 8, 2026
Models change every few weeks. The lasting value is what you build on top: evals, traces, workflows and governance you can carry to the next model.
Adapted from my LinkedIn post of June 17, 2026
Executive Order 14434 renamed AI for federal agencies and kept the legal definition. What it means for a private company, and what it leaves alone.
Adapted from my LinkedIn post of October 19, 2026
Almost every delay between an approved idea and a live user lands in one of seven places. A whiteboard, two dates and the people who waited are enough to find yours.
Adapted from Chapter 3 of The Seven Frictions
The exercise I ask every IT leader to do before anything else. Pick one small use case, stay with it from idea to production, and write down every wait.
Adapted from the introduction to The Seven Frictions