Summary
In this chapter, we explored how to coordinate multiple agents to solve complex tasks by introducing orchestration and handoff strategies in OpenAI Agents SDK.
We first distinguished between deterministic and dynamic orchestration. We then introduced the handoff mechanism, a core OpenAI Agents SDK primitive that enables one agent to pass control to another. We expanded this further with multi-agent switching, allowing agents to transfer control back and forth in longer conversations using persistent memory, and customizing handoffs through callback functions and prompts.
Finally, we explored the four different multi-agent system patterns: centralized, hierarchical (subset of centralized), decentralized, and swarm (subset of decentralized).
In the next chapter, we’ll learn how to manage the underlying workhorse of agents: the LLM.