Published thinking.

The durable decision isn’t which model you use — it’s where, precisely, you place the boundary between the deterministic and stochastic parts of your systems.

The single strongest predictor of whether a task can be safely delegated to AI has nothing to do with how hard it is.

A system does not have one autonomy level. Every action it can take earns its own — classified by reversibility and blast radius.

A frontier lab led with computer-use scores instead of knowledge scores. That editorial choice changes the question you should be asking vendors.

Ten composable lookup tables — maturity, autonomy tiers, oversight modes, verification cost, and action risk classes — designed to be quotable as shorthand.

Before we asked anyone to trust our methodology on their workflows, we ran it on our own tool — and the tool failed its first review.