← All field notes

The org chart of a multi-agent system

Planner, architect, coder, reviewer: why agent topologies end up looking like the teams they replace — and what that means for governance.

Multi-agent systems converge on familiar shapes: a planner decomposes, specialists execute, a reviewer signs off.

Why it matters for governance

When the topology mirrors an org chart, accountability can too. Every agent gets a role charter: what it may read, what it may write, when it must escalate to a human.

Practical defaults

  • One orchestrator, few specialists - resist the 12-agent demo
  • Deterministic guardrails between every hop
  • Human approval gates on irreversible actions
  • Full trace per run, replayable