You are evaluating a customer-support agent’s configuration. Each prospective tool belongs to one of four categories: (1) required to carry out defined tasks, (2) frequently used and reduces hand-offs, (3) occasionally helpful for unrelated work, (4) speculative future utility. Which categories should generally stay in the agent configuration?
Agent tools should be limited to capabilities that support defined tasks and regular operational workflows. Required tools enable the agent’s intended work, and frequently used tools that reduce hand-offs improve its routine execution. Occasionally useful unrelated tools and speculative future tools do not justify current access and unnecessarily broaden the available tool set. OpenAI’s API supports constraining a model to a predefined set of allowed tools.