Single-org versus multi-org multi-agent architecture (SOMA and MOMA)
Key Takeaway
Choose the architecture by where the agents live: SOMA when all agents are in one Salesforce org, and MOMA when multiple Salesforce orgs need to collaborate behind one conversation.
Multi-agent orchestration lets several Agentforce agents work together behind one conversation. The user talks to one orchestrator agent, which acts as the front door. Specialist agents handle the domain work behind it. In Single-Org Multi-Agent (SOMA), all of these agents live in one Salesforce org. Multi-Org Multi-Agent (MOMA) uses the same model, but the orchestrator agent can delegate to specialist agents in other Salesforce orgs.
Why one agent can't scale forever
Beginning in April 2026, agent topics are called subagents. Their function didn't change, so you will see both terms in use.
A single agent can only hold so much. One agent carries too much concurrent intent when it is responsible for more than about 8–10 well-scoped subagents. Its reasoning then starts to drift, get confused or hallucinate.
Hybrid reasoning gives an agent strong reasoning across subagents and deterministic control. However, it does not remove this limit, so one agent cannot scale forever.