The most expensive failure mode in AI-assisted engineering is the agent reporting success without verifying. The fix is one rule: no completion claim without fresh command output.
Verification
Proving the AI's work is correct — TDD, systematic debugging, no-completion-claim-without-evidence.
The most expensive bug in AI-assisted engineering is not the agent writing bad code. It is the agent writing code, claiming it works, and you trusting the claim without checking.
An agent hits a bug, guesses a fix, reports done. Two hours later the same bug reappears because the first fix treated a symptom. Random fixes add bugs. The fix is the iron law: no fixes without root cause.
An agent writes code, then writes a test for it. The test passes immediately. It proves nothing. The fix is one iron law: no production code without a failing test first.