Description

As AI agents move from demos to production, tool design becomes a discipline of its own. Andres and I dig into why treating tools as APIs is the fastest path to reliability, auditability, and multi-tenant safety.

We cover the dispatcher pattern for tool use, how API design changes when agents are the consumers, and how to inject tenant context securely without blowing up your trust boundary. We also share a few hot takes on MCP as we talk about building scalable agent systems in the enterprise.