How we think about AI

A platform prices its auction better than any human can, and a model can carry an expert's method faithfully. Both do their best work when we give them the right inputs and the right limits. Our job is the inputs, not the override.

Numbers, checks, and the act of making a change run as tested code, the same way every time. We reach for AI only where there is no single right answer: which idea, which line, what comes first.

A model writes an analysis in seconds. The decision it points to is the part that still matters, and a person makes it.