Writing

Notes from
the work.

A running record of what I am learning about AI systems, technology leadership, review, and the operating models that hold difficult work together.

Latest writing

These are field notes from a personal project: specific enough to be useful, unfinished enough to stay honest.

From an agentic organization to a work graph

The agents were the easy part. Reliability lived in the handoffs, evidence, rejection routes, recovery, and the point where automation stops.

Separate the doer from the reviewer

Model agreement is not evidence. A review only counts when it can inspect the real record and reverse the answer.

Adopt, borrow, reject

A tool is also an authority model. The useful question is not whether it is good, but what saying yes asks you to hand over.