Orchestrator routing to connected subagents
Key Takeaway
An orchestrator routes to connected subagents based on their descriptions, so give each connected subagent a distinct, specific description. Vague or overlapping descriptions can cause inconsistent routing.
Multi-Agent Orchestration lets one Agentforce agent, called the orchestrator, act as a single front door. The orchestrator routes work to other specialized Agentforce agents in the same Salesforce org, called connected subagents. Multi-Agent Orchestration solves a scaling problem. One agent eventually covers too many domains to route reliably, but many separate agents force users to pick the right one and repeat themselves when they switch.
When one agent stops being enough
Common situations where one agent stops being enough include these:
- A service agent needs to add a field service use case.
- Agents in different business units need to talk to each other.
- A single Instructions block covers many domains, and routing keeps breaking.
Splitting the work into separate agents has its own cost. With Messaging for In-App and Web, users must pick from a growing list of agents and repeat themselves each time they switch. Multi-Agent Orchestration fits these situations, because the orchestrator gives users one front door while each connected subagent owns its own domain.