# When to use a multi-agent architecture

*Agentforce Specialist Study Guide · Multi-Agent Orchestration · Last updated October 15, 2026 · https://www.certifyforce.com/study-guide/agentforce-specialist/multi-agent-orchestration/when-to-use-a-multi-agent-architecture*

When one agent should become several and when it shouldn't: the reliability and control benefits of narrow specialist agents, the supervisor or orchestrator front door, Greeter, Operator and Orchestrator patterns, the four orchestration archetypes (SOMA, MOMA, Salesforce-led and MuleSoft-led multi-vendor), and what Agentforce Multi-Agent Orchestration supports today.

- Certification: Salesforce Agentforce Specialist
- Exam section: Multi-Agent Orchestration
- Study guide: https://www.certifyforce.com/study-guide/agentforce-specialist

**Official exam objective:**

> Given a scenario, determine whether a Multi Agent architecture is appropriate for scalability and control.

## What the exam expects

- Decide whether a requirement needs one agent or several, using reliability, scale and control as the test.
- Cite the benefits Salesforce gives for multi-agent architecture, and the lighter-weight alternatives to consider first.
- Describe the supervisor (orchestrator) agent as a single front door, and tell the Greeter, Operator and Orchestrator patterns apart.
- Choose an orchestration archetype: Single Org, Multiple Agents (SOMA); Multi Org, Multiple Agents (MOMA); multi-vendor A2A led by Salesforce or by MuleSoft.
- Key terms: **monolithic agent** (one agent covering every domain); **supervisor / orchestrator agent** (the front door that routes or delegates to specialist agents); **specialist agent** (a narrowly scoped agent for one domain); **connected subagent** (a complete, separate agent that an orchestrator uses like a subagent, formerly topic); **orchestration archetype** (the system-level blueprint for how a fleet of agents collaborates).

## Key facts

| Fact | Detail | Source |
|---|---|---|
| Default design | Start with a single agent for common use cases | [Enterprise Agentic Architecture and Design Patterns](https://architect.salesforce.com/docs/architect/fundamentals/guide/enterprise-agentic-architecture.html) |
| Reliability principle | Limit any one agent's scope of concerns | [Enterprise Agentic Architecture and Design Patterns](https://architect.salesforce.com/docs/architect/fundamentals/guide/enterprise-agentic-architecture.html) |
| Lighter alternatives | Ensemble of prompts; grounding through actions | [Enterprise Agentic Architecture and Design Patterns](https://architect.salesforce.com/docs/architect/fundamentals/guide/enterprise-agentic-architecture.html) |
| Orchestrator pattern | Stays first point of contact; aggregates specialist responses | [Enterprise Agentic Architecture and Design Patterns](https://architect.salesforce.com/docs/architect/fundamentals/guide/enterprise-agentic-architecture.html) |
| MOMA | Supervisor delegates to another org's agent via A2A | [Enterprise Agentic Architecture and Design Patterns](https://architect.salesforce.com/docs/architect/fundamentals/guide/enterprise-agentic-architecture.html) |
| Handoff vs supervisor mode | Handoff (default): subagent answers directly; supervisor: orchestrator synthesizes | [Multi-Agent Orchestration](https://help.salesforce.com/s/articleView?id=ai.agent_multi_orch.htm&language=en_US&type=5) |
| Feature scope today | One org; no cross-org or non-Agentforce agents; one delegation level | [Considerations and Limitations in Multi-Agent Solutions](https://help.salesforce.com/s/articleView?id=ai.agent_multi_orch_consider.htm&language=en_US&type=5) |
| Timeouts | Orchestrator 120 seconds; connected subagent 30 seconds | [Considerations and Limitations in Multi-Agent Solutions](https://help.salesforce.com/s/articleView?id=ai.agent_multi_orch_consider.htm&language=en_US&type=5) |

## Check yourself

- When should you start with a single agent?
- Why does splitting a large agent improve reliability?
- Which archetype fits when planning starts in a third-party application outside Salesforce?
- When is supervisor mode better than handoff mode?

## Sources

- [Enterprise Agentic Architecture and Design Patterns](https://architect.salesforce.com/docs/architect/fundamentals/guide/enterprise-agentic-architecture.html)
- [Multi-Agent Orchestration](https://help.salesforce.com/s/articleView?id=ai.agent_multi_orch.htm&language=en_US&type=5)
- [Considerations and Limitations in Multi-Agent Solutions](https://help.salesforce.com/s/articleView?id=ai.agent_multi_orch_consider.htm&language=en_US&type=5)
- [Connect an Agent as a Subagent](https://help.salesforce.com/s/articleView?id=ai.agent_multi_orch_connect.htm&language=en_US&type=5)
- [User-Level Authentication in Multi-Agent Solutions](https://help.salesforce.com/s/articleView?id=ai.agent_multi_orch_auth.htm&language=en_US&type=5)

## About the full unit

The "Core concepts", "Exam traps" and "Worked scenario" sections of this unit, its 8 linked practice questions, and unlimited timed mock exams are included with the Agentforce Specialist pack (USD $19): https://www.certifyforce.com/practice-exams