Summary
Cole Medin walks through Google’s Agent CLI — an open-source tool built on top of Google’s ADK (Agent Development Kit) — showing how developers can go from an agent idea to a production deployment without ever opening the framework documentation themselves. The video traces the evolution of agent development over the past two years: from manually writing LangChain code line by line, to wrestling with AI coding assistants that lacked reliable framework context, to today’s workflow where a single CLI install injects all necessary documentation as “skills” directly into the coding agent.
The core concept demonstrated is the pairing of CLI capabilities with skills: the CLI provides commands for scaffolding, evaluating, running, and prompt-optimizing agents, while the skills layer tells the coding agent (here, Claude Code) exactly how to drive each command. Medin builds a code-execution agent live, selects the ADK’s built-in secure sandbox for arbitrary code execution in production, and lets the coding agent handle the entire implementation loop including evaluation iteration.
The video also addresses a common developer question: why use a framework like ADK, Pydantic AI, or LangChain when powerful coding-agent SDKs exist? Medin argues that SDKs excel for personal workflows where output quality matters more than cost and speed, while traditional frameworks remain the right choice for production deployments where token efficiency and response latency are critical constraints.
📺 Source: Cole Medin · Published June 11, 2026
🏷️ Format: Tutorial Demo
