A capable AI agent can do a lot, but some tasks benefit from a team. As in software engineering, dividing responsibilities is only the beginning: the interesting challenges lie in how the parts communicate, share information, and coordinate their work.

This talk explores how agents work together, introducing collaboration patterns before looking at the engineering behind an agent team. How do you define responsibilities? Which state belongs to each agent? And how do you involve people in the collaboration?

To make these concepts concrete, I’ll introduce akgents, our homegrown agent framework inspired by the Actor Model. Its agents maintain their own state and communicate through messages, providing a foundation for teams whose members can work independently while coordinating toward a shared goal. I’ll explain why we chose this approach and demonstrate it with a team of agents that arranges a meeting, finds a room, and organises lunch and drinks.

You’ll leave with a practical understanding of multi-agent collaboration, the architectural choices involved in building agent teams, and a clearer sense of when those teams are worth the additional complexity.