OpenAI’s Jev clone could help the frontier lab stop its swarming agents

OpenAI announced a new Decisions API that functions similarly to a recently released competing product, providing fast and cheap classification capabilities by giving AI models a predefined set of choices to pick from rather than generating open-ended responses. The API could help address a significant problem: monitoring AI agents to prevent misbehavior, since separate models used to watch for bad agent actions currently cost hundreds of dollars per operation but could cost under three dollars using this faster approach. The underlying issue is that traditional large language models are slow and expensive for many software tasks, making faster decision-focused alternatives increasingly attractive to developers. This type of model could theoretically make it feasible to review every single action an AI agent takes, potentially improving overall system reliability.