2025-09-13 –, Track 3
When you let multiple AI agents collaborate, things can get powerful—and messy. How do you decide what each agent should handle? What happens when they step on each other's toes or hit performance walls? This talk dives into those questions, drawing from real-world lessons we learned while building TestZeus Hercules, the first open-source testing agent.
We’ll break down what happens when agents "chat", share tasks, and manage memory together. You’ll hear about the unexpected challenges—like agents clashing over decisions, coordination failures, or struggling with deterministic behavior—and how we tackled them.
This session is for developers who want to go beyond the basics of AI agents and understand their limits. You’ll walk away with practical strategies for setting clear boundaries, improving collaboration between agents, and designing systems that scale without falling apart. If you’re building AI-powered tools or just curious
Outline: Boundaries of Multi-Agent Systems
-
Introduction: Why Boundaries Matter in Multi-Agent Systems (5 minutes)
- Quick story: The unexpected challenges we faced while building TestZeus Hercules
- Why defining boundaries is critical for multi-agent systems
- What attendees can expect to learn from this session -
Breaking Down Multi-Agent Systems: How They Work Together (5 minutes)
- Core components of an AI agent: routers, skills, and memory
- How agents interact and share tasks in a collaborative system
- Real examples from TestZeus Hercules: Coordinating tasks across multiple agents -
Where Things Break: The Real-World Challenges of Agent Collaboration (7 minutes)
- Coordination Failures: When agents clash or overlap on tasks
- Conflicting Goals: Handling disagreements between agents’ outputs
- Resource Limits: Memory, processing power, and execution bottlenecks
- Deterministic Failures: When agents produce inconsistent results -
Setting Boundaries: Defining Clear Roles and Responsibilities (5 minutes)
- How to assign specific tasks to agents and avoid overlap
- Using structured communication to prevent conflicts
- Memory and resource management to keep agents efficient -
Lessons from Hercules: What Worked (and What Didn’t) (5 minutes)
- Key takeaways from developing the open-source TestZeus Hercules
- Code snippets and examples from the Hercules GitHub repo
- Practical solutions we discovered for real-world multi-agent challenges -
Scaling Multi-Agent Systems Without Breaking Them (3 minutes)
- Strategies for building scalable and reliable agent architectures
- Future-proofing your systems with clear design boundaries -
Q&A and Final Thoughts (5 minutes)
- Open floor for developer questions and real-time discussions
- Final thoughts on the future of multi-agent systems and AI-powered collaboration
Key Takeaways for Attendees:
- Understand the core structure and interactions of multi-agent systems
- Learn how to set clear boundaries for tasks, communication, and memory
- Get practical strategies to avoid common coordination and scalability pitfalls
- Explore real examples from TestZeus Hercules and apply lessons to your own projects
Beginner
Robin is the human at the helm for TestZeus, the company behind world's first open source testing agent. He is a versatile engineering leader with more than 15 years of experience in software delivery across startups, scale-ups and enterprises. He is a System 1 thinker, and has led products contributing to $10M ARR. With a metrics-driven approach, he has elevated engineering maturity of product teams for diverse domains such as BFSI, EdTech, Retail, and Developer Experience. Beyond work, he mentors at ADPList and Plato, contributes to open-source projects like Selenium and has authored books and international courses on software testing. He is also a recognized speaker at international events such as Dreamforce (by Salesforce) and Selenium Conference.