Why it matters
NDC AI talk explaining how AI agents use function calling, MCP, and external tools under the hood. It is relevant to agent security because tool schemas, permissions, identity, and execution boundaries define what the model can actually do when natural language turns into actions.
My takeaway: Agent security starts with tool design. Function schemas, MCP servers, identities, approvals, and logs are the control surface that determines whether an agent stays inside its intended scope.