How are teams actually measuring whether docs work for AI agents?

AI-readable docs are already a major conversation.

What still feels underbuilt is the measurement layer.

A lot of teams agree that docs should work for both humans and agents. But in practice, it is still hard to answer questions like:

• can an agent find the right page quickly

• can it parse the structure cleanly

• can it follow setup instructions without guessing

• where does it usually break first
  

That measurement gap is what pushed us to build Docs Benchmark.

I am curious how other teams are thinking about this right now:

- are you measuring agent-usable docs in any systematic way

- what signal do you trust today

- what tends to fail first in real documentation