How do you verify an agent's "this works now" without redoing the work?
Been in a thread lately about the "context tax" of re-explaining project state to an agent every session, and it made me realize the harder problem for me isn't memory, it's trust in the agent's own status reports.
When an agent says "fixed it, tests pass" or "yes that endpoint handles that case," how much do you actually check versus just believing it and moving on? I've gotten burned a couple times by an agent confidently describing behavior it never actually verified, it just pattern-matched to what a fix usually looks like.
Curious what people's actual habits are here. Do you have a personal rule for what always gets manually checked no matter what the agent claims, or has it earned enough trust in your workflow that you mostly just take the report at face value now?
Replies
Ota
This is the problem Ota is built to reduce. An agent saying “fixed” is not enough; the repo should expose an execution record that a human, CI job, or another agent can inspect without repeating the whole investigation.
In an Ota-backed repo, the contract declares setup, runnable tasks, verification workflows, and agent-safe boundaries. ota up, ota run, and ota proof runtime --json produce structured execution and proof output: what Ota selected, what ran, what passed, and what remains unproven.
That does not remove review. It replaces “trust the agent’s narration” with a smaller, checkable question: did the declared verification lane run, against which contract/runtime inputs, and does the execution receipt support the claim?
The broader model is the execution-governance loop: contract, execution, proof, comparison, then policy.
Question: What is the one thing you still manually verify even after an agent reports a green test run?