![[Copilot Studio] Tried Templates, Autonomous Triggers, and Multi-Agent: From Design to Operation Verification](https://images.ctfassets.net/ct0aopd36mqt/wp-thumbnail-00eddef75cb5ee39286e0aa370a96e4f/040bbf127b84da0ee2c483485da1dd29/microsoft-eyecatch.jpg?w=3840&fm=webp)
[Copilot Studio] Tried Templates, Autonomous Triggers, and Multi-Agent: From Design to Operation Verification
This page has been translated by machine translation. View original
Introduction
Hello, I'm Kema.
In the series so far, we have been building up a single agent. Giving it instructions, loading knowledge, and adding behaviors with topics, tools, and flows. This allows you to create "one solid agent."
On the other hand, slightly different requests tend to come up in real-world settings. "Starting from scratch every time is a hassle, so I'd like to start from a template." "I want it to run automatically at set times without a person having to initiate a conversation." "Having one agent do everything gets complicated, so I'd like to split agents by role." These are all about expanding the configuration of agents.
With that in mind, this article covers three topics — templates & managed agents, autonomous triggers, and multi-agent — and summarizes what each can do as of June 2026. We actually ran through creating from a template and multi-agent handoff, and for autonomous triggers we cover the settings screen and confirming prerequisites. I hope this serves as a reference for those who want to broaden their options for building agents in Copilot Studio.
This article is the 4th installment in a series on building agents with Copilot Studio.
Target audience: Those who want to understand how to expand their configurations in Copilot Studio, including using templates, automatic startup, and multi-agent coordination.
Series Article List
| # | Theme | Article |
|---|---|---|
| Part 1 | First Agent | Creating Your First Agent |
| Part 2 | Knowledge | Trying Knowledge-Grounded Answers Based on Files |
| Part 3 | Topics, Tools, and Flows | Building "Behaviors" with Topics, Tools, and Agent Flows |
| Part 4 | Templates, Autonomous Triggers, and Multi-Agent | (This article) |
1. What We'll Do This Time
- Create an agent from a template and actually run it
- Check the types of autonomous triggers and understand the setup and prerequisites for automatic startup
- Organize the three configuration patterns for multi-agent
2. Templates and Managed Agents: Not Starting from Scratch
Available Templates
Agents can be started not just from a blank slate, but also from prepared templates. There are broadly two types.
- Templates: Pre-filled starting points with a name, description, and instructions. You customize these as a base for your own needs.
- Managed Agents: Near-complete solutions built for specific business tasks (many are in preview).
The list for installing managed agents. Business-specific agents are lined up (many are in Preview). Below this same screen, a gallery for "Start with an agent template" continues.
Templates are positioned as "a starting point with content already in place," while managed agents are "near-complete solutions." If there's a template close to what you want to do, you can get to a working state faster than building from scratch.
Verification: Creating and Running from a Template
Whether a template is merely "a skeleton with a name and description" or "something that works right away" isn't clear until you actually build one. So I created an agent from a weather-answering template and ran it.
One thing I noticed during creation is that some templates require a connection to an external connector.
Overview screen of the agent created from the weather template. It comes out with the name, description, and instructions already filled in.
After completing the connection, an agent integrated with the MSN Weather connector was created. I typed "Get Forecast For Today" in the test chat.
When asked about the weather in the test chat, the MSN Weather connector was actually called and a real forecast was returned.
When I asked about the weather, MSN Weather was called and a real forecast came back. This confirmed that templates are not merely "skeletons with names," but are prepared in a ready-to-run state that includes connector integration.
Managed agents and agent templates are currently available in English only and should be limited to internal use within your organization.
Source: Weather（agent template）| Microsoft Learn
3. Autonomous Triggers: Running Without a Person Initiating a Conversation
All the agents up to this point have been "run when a person talks to them."
With autonomous triggers, you can make an agent start running on its own in response to a set event, without a person needing to initiate a conversation.
Types of Triggers
Opening "Add a trigger" from the "Triggers" section on the overview screen displays a list of activation events.
Add a trigger. Various events can be used as triggers, including Schedule (recurring execution), email received, Teams messages, Planner tasks, file updates, and more.
In my environment, the following triggers were available.
- Schedule (recurring execution): Starts at a set time each day, or similar time-based triggers
- Email received (Outlook): Starts when a specific email arrives
- Teams message: Starts when a message comes in on Teams
- Planner tasks / File updates / Various connector events: Starts based on changes in business systems
For example, a process like "every Monday morning, compile last week's KPIs and create a report" can be automated by combining a schedule trigger with the flows from the previous article. It becomes an agent that runs at set times without anyone needing to ask it each time.
Publishing Is a Prerequisite for Autonomous Execution
There is an important point to note here. To actually have the autonomous trigger run automatically, publishing the agent is a prerequisite.
Trigger settings can be configured in the editing screen without publishing. However, the autonomous execution itself — "automatically running when a scheduled time arrives" or "running when an email is received" — only works once the agent is published. The official documentation also states that before publishing, the agent does not automatically react to that trigger.
Before you publish your agent with a new event trigger, the agent doesn't automatically react to that trigger.
Source: Event trigger overview | Microsoft Learn
In this article, we will confirm the types of triggers and their settings, but will not demonstrate autonomous execution, as it requires publishing.
There are also two prerequisites to keep in mind.
- Generative orchestration is required: Event triggers (autonomous triggers) can only be used with agents that have generative orchestration enabled.
- Impact on billing: Enabling event triggers can affect billing (Copilot Credits) calculations. In particular, schedule triggers use more resources the shorter the interval, so care is needed when setting the frequency.
This feature is only available for agents with generative orchestration turned on.
Enabling event triggers can impact how billing is calculated.
Source: Event trigger overview | Microsoft Learn
4. Multi-Agent: Dividing Work Among Multiple Agents
Having one agent do everything causes the instructions, topics, and tools to balloon, making management difficult. Multi-agent is a configuration where agents are divided by role and a parent agent delegates processing to child agents. This time, we'll actually create one child agent and confirm the process of having the parent delegate to it.
Connecting Agents
When you add an agent in the "Agents" section of the overview screen, the "Choose how to extend your agent" dialog opens.
The "Choose how to extend your agent" dialog. Choose from three options: creating a child agent, selecting an agent within the environment, or connecting to an external agent.
There are broadly three types of agents that can be connected.
| Pattern | What to connect | Suitable for |
|---|---|---|
| Child agent | A subordinate agent created exclusively for this agent | When you want to divide a large task by role |
| Another agent in the environment | An existing agent in the same environment | When you want to reuse an existing agent as a component |
| External agent | An agent in a different environment or a published external agent | When integrating across organizations or with external services |
Actually Trying It: Creating a Child Agent and Having It Receive a Handoff
Here we'll create one "child agent." I selected "New child agent" and set the name, Description, and instructions. The Description in particular serves as the basis for the parent's judgment when choosing where to delegate. The key is to clearly describe "what this agent does." For this case, I gave it the role of explaining KPI terminology.
Creating the child agent "KPI Terminology Explainer." The name, Description, and instructions were set and saved. Writing "Explains the meaning of KPI metrics" in the Description lets the parent read this description and decide where to delegate.
After saving, I asked "What does NRR mean?" in the test chat of the parent agent "KPI Report Creation Assistant." The parent did not answer on its own, but instead delegated the processing to the child agent "KPI Terminology Explainer," which returned the meaning of NRR.
When a KPI term is asked in the parent's test chat, the "KPI Terminology Explainer" agent appears in the activity. You can see the input (Task) being passed from parent to child, and the child's output (Summary) being returned. In the end, an explanation covering the difference between NRR and GRR was returned.
The key point is that the user did not specify "use the child agent." The parent read the child agent's Description and determined that this question should be handled by the "KPI Terminology Explainer," then delegated accordingly. This is how multi-agent works with generative orchestration.
Dividing agents by role makes the responsibility of each agent clearer than stuffing everything into one, and also makes maintenance easier. For a KPI report like this one, you could imagine a configuration with a data collection agent, an aggregation agent, and a report creation agent, with the parent delegating to each in sequence.
5. Summary
We confirmed three mechanisms for expanding agent configurations.
- Templates & Managed Agents: A way to start without building from scratch. When we actually created a weather agent from a template, it worked right away including connector integration, and a real forecast was returned.
- Autonomous Triggers: Can automatically start based on events like schedules or email received, without a person initiating a conversation. Trigger types can be checked without publishing, but autonomous execution requires publishing (enabling triggers can affect billing).
- Multi-Agent: A configuration where agents are divided by role and processing is delegated. We actually created the child agent "KPI Terminology Explainer," and confirmed that when a KPI term was asked to the parent, it was delegated to the child and a response was returned. There are three connection patterns: child, in-environment, and external.
Rather than just building up a single agent, knowing how to expand — starting quickly from a template, running automatically, and dividing responsibilities — allows you to choose a configuration suited to your business needs.