Writing · The What if series
Each piece starts with a what-if and ends with evidence — the same habit I bring to systems and teams.
An agent can work out details that nobody scripted in advance. What that flexibility buys, where interpretation and recovery add work, and why the cost needs to be measured against accepted results.
Read the piece →A reviewer can diagnose a failed run and propose better instructions. Frozen candidates, matched comparisons and one owner decision decide whether that proposal is allowed to change anything.
Read the piece →The browser needs the credential. The model needs to know what to do. How private references, controlled execution, and careful handling of observations separate those two needs — and where the boundary ends.
Read the piece →Define what is allowed, check the proposed action, and block it before it happens. A child-facing learning app makes the idea concrete.
Read the piece →Six nodes, twenty-three applications, over a year in operation, one deploy verb. Why running a small company of one is the honest way to earn platform opinions.
Read the piece →A 90-day operating plan written before we have met: listen and map, prove one path, make it repeatable — and four things I will refuse to do along the way.
Read the piece →Why I built this site's visitor counting myself: privacy by design, rules that live in the system, and lessons from a one-day build that carry to enterprise AI.
Read the piece →Continue the conversation