Do you review the AI's tests before reading its code?

It’s been a couple weeks now since I’ve begun doing it the other way around. Instead of inspecting the function itself, I begin by reading the tests that the agent generated for it. These tests reflect the agent’s understanding of what the code is supposed to do, which is not necessarily what I asked for.

Twice now I found an oversight in ten seconds that otherwise would take an hour of debugging. The code was fine both times. The tests pointed the issue out.

Are there any other people reviewing code in this manner? And how do you decide whose code is the true one when tests themselves are automatically generated?