{}Route To AI
Roadmaps
← All roadmaps
Advanced

Design and run multi-agent systems

Know when one agent is enough, and how to run many safely.

For: Engineers and tech leads who already built a basic agent.

4 steps
stops on the route
12 lessons
you finish in total
2 courses
the lessons come from
~4h reading
all lessons together
Adds it to My learning so you can track it.
AI careerThis is the roadmap for the Agentic orchestrator role.
Read about the role →

Finish every lesson to earn a certificate of completion that anyone can verify. Jump to the certificate ↓

Other roles:Agentic engineerForward deployed engineerAI transformation strategistAI evaluation engineerAI security engineerAI governance and risk leadAI product manager
Your progress on this roadmap0 / 12 lessons completed0%

How to read this page. A roadmap is a route made of 4 steps. Each step has one or more lessons. There are 12 lessons in total, taken from 2 courses. Extra reading is listed at the end of the page. It does not count towards your progress.

Certificate of completion

Earn your Agentic orchestrator certificate

Verified edition

Mark every lesson in this roadmap complete. You have finished 0 of 12. 12 to go.

See a sample certificate. Anyone can check a certificate with its code at /verify. It records course completion. It is not an exam or an accredited qualification.

  1. 1

    Step 1 of 4

    Decide if you need many agents

    Tell when a single agent is the better choice.

    Lessons0/2 done

    Before you move on: Take one idea and argue for the simplest design that could work.

  2. 2

    Step 2 of 4

    Split work and pass it on

    Break a task into parts and hand context between agents without losing it.

    Lessons0/3 done
    Tool: LangChain & LangGraphTool: CrewAITool: Agent Development Kit (ADK)

    Before you move on: Write down exactly what each agent receives and returns.

  3. 3

    Step 3 of 4

    Make it reliable

    Plan for failures, cost and speed from the start.

    Lessons0/3 done
    Tool: Model Context Protocol (MCP)Tool: OpenTelemetry

    Before you move on: List five ways your system can fail and what happens in each case.

  4. 4

    Step 4 of 4

    Run it in production

    Watch, test, approve and govern agents once they are live.

    Lessons0/4 done
    Prompts: AI agent promptsTool: LangfuseTool: Arize PhoenixTool: Helicone

    Before you move on: Complete the production-readiness review for a system you know.

Hands-on lab · build confidence

Run a three-agent report pipeline and break it on purpose

A researcher, a writer and a reviewer pass work to each other. You add limits and test the failures.

No code · 3 to 5 hours

Start the lab →Practice interview questions

Other roadmaps

Use AI well at work

Get reliable results from AI in your daily job, with no coding.

Become an AI agent engineer

From how models work to agents you can ship and trust.

Design and ship websites with AI

Go from a brief to a live, good-looking site, and work with clients.

Create images, 3D and video with AI

Make a consistent, usable visual style, and know the licensing rules.

0 / 12 finished0%