Hierarchical Delegation
Hierarchical delegation pattern for multi-agent systems: orchestrating specialized agents with a coordinator for complex tasks.
Last updated
After this section you can
- Design the delegation contract between an orchestrator and its specialists — the real API of a multi-agent system
- Tell orchestrator, planner, and router apart instead of conflating them
- Recognise the coordination failure modes and the anti-patterns that cause them
Hierarchical Delegation: Multi-Agent Teams
An orchestrator agent decomposes the mission and delegates to specialists — each with its own context window, tools, and model tier. The pattern behind Claude’s Research mode, and the most expensive way to be right.
A single agent eventually hits three ceilings: the context ceiling (one window must hold research, code, and review at once), the focus ceiling (a prompt juggling six roles does all of them worse), and the tool ceiling (forty tool schemas confuse selection). Hierarchical delegation splits the job like an engineering org: an orchestrator owns the mission, decomposition, and synthesis; specialists own one competency each, with only the tools and context that competency needs. The win is not intelligence — it is isolation. The cost is coordination, and it is never free.