Improving your agents is like gradient descent

Improving your agents is sort of like gradient descent, where you take small steps and keep whatever gets you closer. Try improving things through AGENTS.md, the reasoning level and your tooling, and find which changes get you fewer errors.