Supervisor Pattern

• Key points are in the visual diagram above.

Complex workflows need coordination — who scrapes the web, who writes code, who checks compliance? A supervisor maintains global context and decides which worker to invoke next, enabling multi-agent systems without chaos.

A supervisor is a project manager who assigns tasks to team members, reviews deliverables, and compiles the final presentation — without doing every job themselves.

Scenario

User: 'Build a landing page for our new product.' Supervisor → researcher (gathers product info) → designer agent (layout suggestions) → coder agent (generates HTML) → supervisor reviews and returns final page.

Solution

In Agent Design Patterns, apply Supervisor Pattern to this scenario: User: 'Build a landing page for our new product. 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 Supervisor Pattern (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 Supervisor Pattern happens in the code.

1from langgraph.prebuilt import create_react_agent  # import dependencies2
3researcher = create_react_agent(llm, [web_search, scrape])4coder = create_react_agent(llm, [write_file, run_code])5
6def supervisor(state):  # define a reusable function7    decision = llm.invoke(f"Delegate next step. Workers: researcher, coder\n{state}")8    if decision.worker == "researcher":9        return researcher.invoke(state)  # return the result10    return coder.invoke(state)  # return the result

• Treating Supervisor Pattern as a black box without evaluation
• Ignoring cost and latency in production
• Skipping error handling for supervisor pattern

Quick recap — the most important points from this module.