Your AI Coding Agent Doesn’t Need Another Prompt. It Needs a Judge.

For the last couple of years, we have spent an incredible amount of time trying to make AI coding agents smarter. Better prompts. Longer context windows. More capable models. More tools. More

For the last couple of years, we have spent an incredible amount of time trying to make AI coding agents smarter. Better prompts. Longer context windows. More capable models. More tools. More autonomy. All of that helps. But a recent developer experiment convinced me that we may be focusing too much on the agent and not enough on the system that tells the agent when it is wrong. Developer Maurice Heumann spent roughly three months using multiple AI agents to reconstruct a large first-person shoo