Subagents versus a Multi Agent architecture
Key Takeaway
When one agent's Instructions block covers many domains and routing keeps breaking, or separate specialized agents need to work together, add the specialized agents to an orchestrator agent as connected subagents instead of growing one agent. Users get one front door, and each connected subagent stays its own agent.
An Agentforce agent splits its work into subagents, and each subagent holds the instructions and actions for one job. When one agent outgrows that structure, Multi-Agent Orchestration lets an orchestrator agent route work to connected subagents, which are complete, separate agents in the same org. Knowing the difference tells you whether to add a subagent to one agent or to connect separate specialized agents.
Subagents inside one agent
Beginning in April 2026, agent topics are called subagents. How they work did not change, so you may still see both terms.
When you create an agent, Salesforce adds a special subagent called the Agent Router. The Agent Router is the starting subagent for each conversation.
By default, the agent uses the Agent Router for subagent classification. The agent compares recent conversation history, including the user's latest message, to the names and descriptions of all subagents and picks the best match.
You can edit the Agent Router like any other subagent to control routing inside one agent. You can add conditions, actions and variables, or use filters to exclude a subagent. You can also remove references to a subagent so that the subagent is reachable only through transitions from other subagents. For example, a filter on a verified variable can keep an Order Management subagent available only to verified customers and send everyone else to a verification subagent.