07_delegate

07 — Delegate (Multi-Agent Coordination)

One agent calls others as tools — research, write, done.

What this shows

• mode: delegate in orchestrations: — turns sub-agents into callable tools
• entry_name: — explicitly names the coordinator agent whose blueprint is used
• connections: — a flat list of child agents, each with a description: the coordinator sees when deciding which tool to call
• The simplest multi-agent pattern: a coordinator that delegates, never writes itself

How it works

Define your agents normally, then wire them with an orchestration:

orchestrations:
  main:
    mode: delegate
    entry_name: coordinator
    connections:
      - agent: researcher
        description: "Research a topic and return structured facts."
      - agent: writer
        description: "Write a report from the provided facts."

entry: points at the orchestration name (main), which is all you need in main.py.

When load() resolves the orchestration, it forks a brand-new Agent from the coordinator blueprint (model, system prompt, hooks, kwargs — everything). researcher and writer are each wrapped as async @tool functions and added to that new agent. The original coordinator agent defined under agents: is never mutated.

coordinator blueprint
  └── forked into new Agent
        ├── researcher (tool)  — gathers facts
        └── writer     (tool)  — turns facts into a report

The forked agent's agent_id is the orchestration name (main), making it easy to trace in logs. Any agent_kwargs defined on the orchestration are merged on top of the coordinator's own kwargs — orchestration values win on conflict.

Good to know

The fork, not the original. The agent you interact with via resolved.entry is the newly built fork, not the coordinator you declared. This matters if you pass the same agent into multiple orchestrations — each one gets its own independent copy.

Each sub-agent is independent. It gets its own model, system prompt, tools, and hooks. The coordinator (fork) only sees the description: string — it can't peek inside.

The coordinator should not write content itself. Its job is to call the right sub-agent at the right time and pass through the result. The system prompt enforces this.

Prerequisites

• AWS credentials configured (aws configure or environment variables)
• Dependencies installed: uv sync

Run

uv run python examples/07_delegate/main.py

Try these prompts

• Write a short guide on how Python manages memory.
• Create a report on the benefits of test-driven development.
• Produce a concise overview of how HTTP/2 improves on HTTP/1.1.

Advanced topic — suppress default callback logging

Strands agents log actions to the console through their default callback_handler. If you want cleaner example output, set the handler to null in agent_kwargs for any agent:

agents:
  my_agent:
    agent_kwargs:
      callback_handler: null # or ~