Design multi-agent solutions in Copilot Studio
Unlock the complete study guide + 1,040 practice questions across 16 full exams.
Bundled into the existing Designing and Building Integrated AI Agent Solutions in Copilot Studio premium course — no separate purchase.
14-day money-back guarantee — no questions asked.
Included in this chapter:
- Where the multi-agent decision actually lands
- The signal to split, and the numbers that are not it
- Child agents: one deployment, its own orchestration
- What the main agent does when a child agent finishes
- Connected agents: independent lifecycle, local description
- Choosing the coordination pattern
- Choosing the connection option by where the agent was built
- What actually crosses on every delegation
- Instruction design so exactly one agent speaks
- Explicit call sites: topic redirect and instructions
- Governing delegation across the trust boundary
- Testing a multi-agent design before you ship it
- How these choices show up in questions
Where a capability lives, and what that choice costs
| Consideration | Single agent | Child agent | Connected agent |
|---|---|---|---|
| Where it is defined | Topics, tools and knowledge on one agent | Inside the main agent, on its Agents page | A separately published agent the main agent connects to |
| Who owns and releases it | One team, one release | The same team that owns the main agent; deploys with it | Can be a different team, with its own ALM process |
| Orchestration and tool budget | One orchestration; 128 tools maximum, 25 to 30 recommended | Its own orchestration and its own set of up to 128 tools | Its own orchestration, its own tools and knowledge |
| Conversation context | Single context throughout | Shares the main agent's context | Conversation history passed by default; can be cleared |
| Added latency | None beyond the single orchestration | An added layer of orchestration | Orchestration hops through a separate orchestration layer |
| Reuse by other agents | Not reusable as an agent | Not published separately or reusable elsewhere | Usable by multiple main agents, if it has no connected agents itself |
| Transcripts and governance | One set of transcripts and policies | Activity appears in the same session as the main agent | Separate transcripts and policies; sessions must be correlated |
| The signal that says choose this | Single intent or task, one cohesive team, no reuse | Logical grouping inside one solution, no separate publication | Independent ownership, settings, publication or reuse |
Decision tree
Cheat sheet
Unlock with Premium — includes all practice exams and the complete study guide.
Also tested in
References
- Add a child agent - Microsoft Copilot Studio
- Orchestrate agent behavior with generative AI
- Add other agents overview - Microsoft Copilot Studio
- Understand multi-agent architectures in Copilot Studio
- Add tools to custom agents (limitations on tools in agents)
- Design multi-agent solutions using child agents
- Connect to an existing Copilot Studio agent
- Orchestrator and subagent multi-agent patterns
- Design multi-agent solutions using connected agents
- Multi-agent patterns (agent architecture guidance)
- Multi-agent orchestration patterns and best practices
- Apply generative orchestration capabilities