Agent Orchestration

Demo

Click to play with sound.

---
name: Agent Orchestration
description: Design reliable multi-agent systems.
---

# Agent Orchestration

Coordinate multiple LLM agents to solve tasks a single prompt can't.

## When to use multiple agents

- The task decomposes into distinct skills (research, then write, then review).
- Context would overflow a single agent, so you split and summarize.
- You need parallelism across independent subtasks.

Don't multiply agents needlessly — each hop adds latency, cost, and failure surface. Prefer a single capable agent until it clearly can't cope.

## Topologies

- Orchestrator-workers: a planner decomposes the task and routes subtasks to specialized workers, then synthesizes results.
- Sequential pipeline: fixed stages, output of one feeds the next.
- Router: classify the input, dispatch to the right specialist.

## Task routing
… install to load the full skill