What If AI Coding Agents Could Predict a Bad Code Change Before Making It?

AI coding agents usually follow this loop: Make a change → Run tests → Fix errors → Try again. But...