The newsletter

Where the last six weeks went, and what I decided to build instead: a monthly case note series on the economics of AI in production, built from the filings. The first one is Klarna.

Everyone says AI turns experts into managers. It does the opposite. The depth has to sit with the reviewer now, because there is nowhere else for it to sit.

The Agent Estate Standard is published. Every requirement in it can be met with a document, except one. Requirement 7.3.7 has to have happened.

The number every budget is built on has stopped measuring what everyone assumes it measures.

The week-one sequencing decision that decides your AI programme: governance, security and change as streams, not phases.

Why CAIOs are getting the org chart right and the unit economics wrong.

Five structural conditions doing the work of producing failure, named from six issues of practitioner replies.

The portfolio you have not counted.

Why retraining your AI keeps failing, and what replaces it.

When cheaper tokens make the problem worse: the condition where every task an agent completes costs more than the human equivalent.

The failure mode no dashboard is showing you.

Why the perfect pilot is the most dangerous one to scale.

Why the world’s best firms cannot agree on what “working” means for AI agents.