What's this about?
Cognition has Devin test its work, record the application running and report both passed and untested checks. It can also return a screenshot after fixing a screenshot-reported bug.
What exactly is acceptance evidence?
It is observable proof that a change behaves as requested—recordings, screenshots and test results—not merely the agent saying “done.”
Why is it worth attention?
Verification is becoming the engineering bottleneck as agents write more code. Evidence improves review efficiency, but humans must still judge coverage and whether the demonstration matches the requirement.