Why AI browser agents are slow, measured step by step

Where a real run spends its time: deep-model escalations, page observation, and the model round trips that turn out to be the smallest part. What a replay cache changes, and where it did not help.

The missing piece for AI browser agents

Why browser agents feel slow, what Jev (TypeSafe's System One model) changes about each decision, and why making it work on every website is the hard part.