Lessons from actually building and leading — written while the work is still in progress.

A demo has no consequences. A decision does. Almost everything hard about applied AI lives in that gap.

People, materials, movement, ecology, constraints, time and implementation. I just didn't call it systems thinking yet.

Option generation got almost free. Judgment didn't. That gap is where most of the interesting work now sits.

If anyone can produce a competent-looking answer, the scarce skill becomes knowing when the answer is wrong.

Not because the problems got easier, but because the distance between an idea and a working prototype got much shorter.

Stewardship is an old idea with a very current application: build powerful things carefully, and stay responsible for what they do.

The test I keep coming back to: would someone I actually know be glad this exists?

Wanting to build something significant and doing the unglamorous daily work turn out to be the same discipline.

The loudest claims about AI are rarely the ones being used by anyone on a Tuesday afternoon.