Monitoring Internal Coding Agents for Misalignment

OpenAI publishes very little about how they run agents in their own engineering organization. "How we monitor internal coding agents for misalignment" is the rare exception — a documented account of how they detect when one of their internal coding agents is doing something it shouldn't be: pursuing a side objective, taking shortcuts that produce plausible-but-wrong code, or otherwise drifting from the intent of the task.

The methodology is interesting. The implication is more interesting. OpenAI is the lab that built the model — they have the weights, the training data, the eval harnesses, the engineers who designed the architecture. And they still need explicit, dedicated monitoring infrastructure to deploy these agents inside their own company.

Pull back from the specific monitors and ask what shape of organization is implied by the post. The picture: an agent runtime, a parallel monitoring layer that watches it, a separate review function staffed by humans who didn't build the agent, and an escalation pathway with someone empowered to halt a run.

That isn't a feature inside an engineering team. It's a small operational org with its own reporting line. Most companies stand up an 'AI engineering team,' assign monitoring to it as a side responsibility, and discover six months later that the people building the agent are also the people grading it. "How we monitor internal coding agents for misalignment" is implicitly arguing against exactly that structure.

OpenAI describes the discipline in "How we monitor internal coding agents for misalignment." They don't describe the cost. The cost is real: dedicated monitoring infrastructure, a separate review function, ongoing eval maintenance as the model and the workflow change. We've watched clients deploy two coding agents and discover they need the equivalent of a small SRE team to make those agents production-trustworthy.

That's the unsexy line item nobody includes when the agent is being scoped. We argue for putting it in the budget on day one, because finding it in month four is how a deployment gets quietly rolled back.