What counts as proof that an agent tool works?
“The code exists”, “the tests pass”, “it’s installed”, “it ran for real” and “a person used it” are different claims. We say which one we have.
Passing tests and a person actually getting the job done are both good signs. They aren’t the same proof.
Six levels of proof
Each level proves something different:
- 1Written
The code exists.
- 2Tested
Automated checks pass on the cases we wrote.
- 3Visible
A running agent can find the tool.
- 4Installed
An exact version is running, healthy, on a real server.
- 5Ran for real
One real job went all the way through, past every check it should.
- 6Used by a person
Someone did the work the tool was built for.
Blur these together and you get impressive claims nobody can repeat. Keep them apart and every gap tells you what to do next.
Test the job that matters
An agent that can see a tool hasn’t proven that the tool’s most important action works. A listed connection to another service hasn’t proven the login, the right account, the permissions or the handling of replies. A test with sample data hasn’t proven the client’s real setup. A proper sign-off record names:
- the exact code version;
- how it was built and checked;
- which version went out;
- whether the service is up and passing its health check;
- one real job done end to end;
- any check a person still has to do.
Say what isn’t proven yet
Anyone evaluating agent software needs to know what they are looking at: code that’s written, a version that’s installed, or a workflow a person has already signed off. Notes on proof belong next to the sales claims, in plain words, not buried in engineering language. This doesn’t make the work sound smaller. It makes progress easy to read and gives every open gap a clear next proof.