Task Delegation

Dumping entire chat logs into worker prompts wastes tokens and confuses specialists. Smart delegation transfers intent, constraints, and relevant artifacts only.

Delegation is writing a clear Jira ticket for a colleague — acceptance criteria, context links, deadline — not forwarding a 200-message Slack thread.

Scenario

Supervisor delegates `{task: 'Summarize attachment', file_url: '...', max_words: 300}` to Document Agent — no billing context included.

Solution

In Multi-Agent Systems, apply Task Delegation to this scenario: Supervisor delegates `{task: 'Summarize attachment', file_url: '. Identify the inputs, run the technique, validate the output, and note one thing you would monitor in production.

Do this before moving to the next module — reading alone is not enough.

Open the Code Walkthrough below and run it locally. Change one parameter related to Task Delegation (e.g. model, temperature, top_k, or tool name), observe the difference in output, and write 2–3 sentences explaining what changed.

Highlighted lines show where Task Delegation happens in the code.

1@tool2def delegate_research(query: str, max_sources: int = 5) -> str:  # define a reusable function3    result = research_agent.run(ResearchTask(query=query, max_sources=max_sources))  # key line for Task Delegation4    return result.summary  # return the result

• Treating Task Delegation as a black box without evaluation
• Ignoring cost and latency in production
• Skipping error handling for task delegation

Quick recap — the most important points from this module.