The Frontier: where measurement has to arrive before the stack does

Most AI value thinking assumes a human at a keyboard judging whether the tool helped. That assumption is expiring. These three essays follow value measurement into the territory where it gets hard: agents with no user to survey, AI that builds AI, and models that leave the metered cloud entirely.

This is the most original ground on the hub and the most exposed, so the speculation tags do heavy lifting here. That is deliberate.