You're building an LLM agent that calls backend tools to complete user requests across many product surfaces. As the number of tools grows, schemas drift, APIs change, and bad tool calls start showing up in production.
Walk me through function calling at scale. How do you keep tool definitions in sync with the actual APIs?