Loop

Multi-Agent Orchestration

Coordinating many Claudes: research-system lessons, the SDK loop, and long-horizon state.

0 of 7 lessons complete · 94 min total

Start module
  1. Lessons from Anthropic's Research SystemWhat building a production multi-agent researcher taught: effort scaling, delegation clarity, tool design, and why parallel subagents beat one long context for breadth-first work.Lesson · 16 minNot completed
  2. Orchestrator + Subagents + SynthesisThe full multi-agent research architecture animated.Animation · 10 minNot completed
  3. The Claude Agent SDK LoopThe gather-context / take-action / verify loop the SDK implements, and how tools, permissions, and hooks plug into it.Lesson · 15 minNot completed
  4. Long-Horizon AgentsKeeping state across context windows: compaction, external notes and memory files, and structured handoffs between sessions.Lesson · 15 minNot completed
  5. Choosing an Orchestration PatternA decision framework: task decomposability, verification cost, latency budget, and when a single agent with good tools beats a team.Lesson · 15 minNot completed
  6. Multi-Claude Workflows in PracticeWriter/reviewer splits, git worktrees for parallel agents, and agent teams with frozen contracts and disjoint file ownership.Lesson · 14 minNot completed
  7. Multi-Agent CheckpointOrchestration trade-offs, state strategies, and parallelism failure modes.Checkpoint · 9 minNot completed