Copilot (licence) - Configure a child agent Tutorial

Explore how to configure a child agent and orchestrate a coordinated service center. Learn to name the agent, set call conditions, and trigger rules such as message reception, activity types, conversation updates, and inactivity handoffs. See how inputs/outputs can be customized, including date extraction and advanced validation, to create a truly intelligent, well-coordinated assistant. A practical guide to expanding capabilities via inter-agent connections and precise input handling.

03:52
152 views
Parteger

Available in AI - Artificial Intelligence

00:00:09

Imagine a technical support agent performing triage,

00:00:12

redirecting the user towards specialized agents.

00:00:16

By clicking add,

00:00:18

you can connect to an agent already available in your environment,

00:00:22

provided this option has been enabled in the settings.

00:00:26

Or to external agents through Microsoft Fabric,

00:00:29

Microsoft Foundry,

00:00:31

Microsoft 365 agents SDK,

00:00:34

or the agent to agent protocol.

00:00:38

We will now create a child agent.

00:00:41

Just like a general assistant,

00:00:43

you must configure a name,

00:00:45

but above all define the situations in which

00:00:48

the main agent will call this child agent.

00:00:51

Assistant chooses

00:00:53

triggered based on the description of the child agent.

00:00:57

A message is received,

00:00:59

triggered when a message activity,

00:01:01

the most common type,

00:01:03

is received.

00:01:04

A custom client event occurs,

00:01:07

triggered when an event type activity is received.

00:01:11

An activity occurs,

00:01:13

triggered when any type of activity is received.

00:01:16

The conversation changes

00:01:19

triggered when a conversation update activity is received.

00:01:25

It's invoked,

00:01:26

triggered when a calling activity is received.

00:01:30

It is redirected to

00:01:32

triggered when the assistant is explicitly called from a topic.

00:01:36

The user is inactive for a while,

00:01:39

the agent takes over if no interaction is detected after a defined period.

00:01:44

A plan completes

00:01:46

triggered when the main assistant finishes executing all planned

00:01:50

steps for a user request or an autonomous trigger.

00:01:55

An AI generated response is about to be sent

00:01:59

triggered when the assistant generates the final message intended for the user

00:02:03

after consulting the necessary topics,

00:02:06

tools,

00:02:06

or knowledge sources.

00:02:09

Each criterion also includes advanced settings

00:02:13

such as conditions and prioritization.

00:02:17

Most other criteria are identical to those used for the assistance instructions,

00:02:22

knowledge sources,

00:02:23

or tools.

00:02:25

What makes a child agent different is the inputs and completion.

00:02:30

By default,

00:02:31

it receives a natural language task from the main agent.

00:02:35

Once its mission is completed,

00:02:37

it returns a natural language summary of what it accomplished.

00:02:42

However,

00:02:42

you can specify more precise input and output data.

00:02:48

For example,

00:02:49

if your child agent requires a date,

00:02:51

and the main agent has not provided it,

00:02:54

the child agent will ask the user directly to identify it as an entity.

00:03:03

In the advanced section you can customize the interaction,

00:03:07

the number of attempts to request information,

00:03:11

whether additional validation is needed,

00:03:13

or the procedure to follow if no valid data is found.

00:03:18

Completion allows you to choose what the agent outputs.

00:03:23

In our example,

00:03:24

it can extract a date from the corresponding entity.

00:03:28

Once completed,

00:03:30

simply save the child agent to make it available.

00:03:34

The list of assistants will then be displayed,

00:03:37

and they can be modified or deactivated later.

00:03:41

Child agent orchestration is the key to transforming your assistant

00:03:46

into a truly intelligent

00:03:48

and perfectly coordinated service center.

Mandarine AI: WHAT YOU SHOULD KNOW

RELATED COURSES This tutorial is part of the following courses

E-learning Débutant

Vu 405 fois

Get Familiar with Copilot Studio

Build and deploy custom AI assistants in Microsoft Teams with Copilot Studio. Optimize interaction with your teams with a simple interface and features tailored to your collaborative needs.

ALL COURSES