TL;DR: A senior Anthropic engineer published a free one-hour walkthrough of building coordinated Claude agents using loops and graphs, covering CLAUDE.md, plan mode, skills, hooks, and subagents.
Summary: A senior engineer at Anthropic released a one-hour course on constructing teams of cooperating Claude agents. The curriculum moves from CLAUDE.md configuration and plan mode, through building reusable skills and hooks, into multi-agent and subagent orchestration, and finishes with self-improving loops and graphs. It packages Anthropic's internal-style agent engineering patterns into a single condensed walkthrough.
Why it matters: It's a rare first-party-flavored reference for how Anthropic practitioners actually structure agent loops, tool hooks, and subagent delegation rather than ad-hoc prompting. Worth an hour if you're designing multi-agent systems, and worth comparing against your current orchestration approach for the loop/graph patterns.
Source: x_com