Everyone's building multi-agent systems, but nobody agrees on how. This talk proposes a taxonomy of five frontier multi-agent strategies and shows what happens when you compose them into a single architecture. Drawing from production data at Factory, we walk through a three-role system (orchestrator, workers, validators) that uses validation contracts, structured agent handoffs, and adversarial verification. We cover the case for serial over parallel execution, why model selection per role is a compounding advantage, and how to design systems that get better with each model generation instead of being made obsolete by them.

Speaker info:
- https://github.com/lukealvoeiro
- https://www.linkedin.com/in/lukealvoeiro

AI Engineer

Luke Alvoeiro from Factory presents a multi-agent architecture called 'Missions' designed to run software development tasks autonomously for days or weeks. The core argument is that the bottleneck in modern software engineering is human attention, not intelligence. Missions uses a three-role architecture: orchestrators (planning), workers (implementation), and validators (verification). Key design decisions include serial feature execution to avoid agent conflicts, validation contracts written before any code to define correctness independently of implementation, structured handoffs between agents to preserve context, and model-agnostic design so the right LLM can be placed in each role. The system combines five multi-agent communication patterns: delegation, creator-verifier, broadcast, negotiation, and direct communication. Production results show missions running up to 16 days, with 50% of final code being tests and 90% test coverage. The architecture is prompt-driven rather than hard-coded to improve automatically as models improve.

Multi-Agent Systems That Ship for Days — Luke Alvoeiro, Factory