Agentic Systems
Design and build AI systems where several agents work together.
Design and Build Multi-Agent Systems
When one agent is not enough, the hard part moves to the gaps between agents: who does what, how work is handed off, and what happens when one of them gets it wrong. Work with Sunil Mathew to design, build and evaluate a multi-agent system, and explain its architecture.
| Programme overview | Details |
|---|---|
| Who this is for | Senior engineers, technical leads and architects whose problem needs more than one agent. |
| Learning outcome | Build a working multi-agent system, explain its architecture and evaluate where it holds and where it breaks. |
| Duration | 5 weeks |
| Format | Live cohort learning, with 3–4 hours of independent work a week in addition to live sessions. |
| Cohort | A live cohort of 8, never more than 10. It starts once 8 participants have confirmed. Sessions are not recorded. |
| Certificate | A certificate of completion from Sunil Mathew. It is not accredited by a university or certification body. |
| Hours | 20 live cohort hours |
| Fee | $999 per participant |
| GST | Not charged at the moment. If that changes, we will tell you before you pay. |
| Next step | Request a fit call |
Start with the right question
Do you need several agents, or one agent with better tools? You learn to make that call before you build, and to show the evidence for it.
“Is this the right use case for an agent at all? Many problems are better solved with ordinary software.” — Sunil Mathew
Where multi-agent systems break
The difficult engineering sits in the gaps between agents. These are the hard cases you build for and test.
- A failed handoff: work is lost between agents.
- Conflicting outputs: two agents produce results that disagree.
- A loop: an agent keeps going instead of finishing.
From use case to working system
-
Decide the shape
Bound the problem and agree what success looks like. Decide whether it needs one agent or several, and why.
-
Design the roles and handoffs
Give each agent a clear responsibility, tool access and limits. Define how work and context pass between them, and where a person reviews.
-
Build and evaluate
Build the system, then test it with the hard cases: a handoff that fails, conflicting outputs, an agent that loops. Read the traces and costs.
-
Revise and explain
Use the evidence to revise the design. Explain your architecture and its remaining risks as you would in a design review.
What you work towards
A working multi-agent system, an explained architecture and an evaluation of where it holds and where it breaks. The remaining engineering work becomes part of the result you can discuss.
The system is a learning project. Organisational deployment requires its own work on data, security, access, reliability and operating responsibility.
Fit and commitment
For senior engineers, technical leads and architects whose problem needs more than one agent. Discuss your current capability and proposed scope before joining.
- 20 live cohort hours with Sunil.
- 3–4 hours per week of independent building and testing.
- $999 per participant, with schedule and payment arrangements discussed before joining.
Review the reasoning with an experienced engineering leader
Sunil Mathew has held engineering leadership roles at Google, Amazon and Walmart Labs across a 26-year career. His current work includes agentic-AI products, multi-agent orchestration and teaching. The live discussions examine why your system behaves as it does and what evidence should guide the next change.
Comparing your options
Building your first agent? See Build Reliable AI Agents. Leading the design, review and release of a whole agentic system? See Architecting Agentic AI Systems.