The building blocks of an AI agent

This article outlines the five-layer architecture of production AI agents: user interface, agent framework, tools/MCP servers, LLM/model, and auth/identity. The piece emphasizes that authentication is not a feature added at the end but rather a foundational concern that threads through every layer of an agentic system, including human-in-the-loop approvals, fine-grained authorization, and comprehensive audit trails.