[tonyww]

A useful way to think about this: browser-use is the runner, Sentience is the assertion layer.

Agents act → Sentience verifies.

If you’ve built flaky E2E tests or agent demos that “usually work”, this is an attempt to make those workflows inspectable and testable.