OpenAI said it is becoming increasingly important to evaluate the performance of AI agents in “economically meaningful environments” as their adoption grows.