The situation
The agent says the feature is complete. The code looks convincing and the explanation names a test. But nobody ran the test, opened the page or checked the deployed URL. "Done" is still a claim.
The idea in one paragraph
Verification is evidence matched to the claim. A typecheck can show that types are consistent; it cannot show that a mobile menu is usable. A browser check can show that one flow works; it cannot prove a backup can be restored. Choose the cheapest independent check that could genuinely expose the failure you care about.
| Claim | Useful evidence | Still unproven |
|---|---|---|
| It builds | A fresh production build exits zero | The page is readable |
| The flow works | Walk it in the browser | Other data and devices work |
| The route is live | Public request returns expected body | Every edge location is healthy |
| It is recoverable | Restore drill succeeds | Future backups will run |
How it actually works
Start from the claim, then name a check with a different failure path. Reading the same code that produced the bug is weaker than running it. Accepting the agent's summary of its own work is weaker than inspecting the diff or output. High-risk actions need a higher evidence bar before they happen, not a more confident explanation afterward.
Secrets, destructive commands, publishing, money, personal data and external messages are hard to undo. Resolve the exact target, prefer a reversible step, keep sensitive values out of output and confirm the destination before information leaves your machine.
What this changes for you
- Ask what evidence would make the main claim false, then run that check.
- Match verification to the surface: tests for logic, browser for interaction, public request for deploy.
- Slow down at hard-to-undo boundaries and identify a rollback before acting.
Where it breaks
No practical check proves everything. Tests can encode the wrong expectation, a green build can hide runtime failure and a manual check can miss an edge case. Verification reduces uncertainty; it does not erase it. Increase depth with consequence, not with anxiety.
Terms used on this page
- Independent verification: evidence gathered through a different path from the claim itself.
- External side effect: an action that changes another system or affects another person.
- Reversible action: a change with a known, practical way back.
- Rollback: the prepared path for restoring the last known good state.
Read next
Level 2: Tools and data maps the systems around the agent and the project you just learned to direct.