Creating Copilot Studio agents (chatbots or virtual agents)

This page has 0 threads | Add post

Blogs

Finding our blogs useful? Please consider helping us to share more!

Explaining the two different types of Copilot Studio agent Part three of a three-part series of blogs
Copilot allows you to create chat bots called "agents", but these can be either simple conversational tools or sophisticated bots depending on which type of Copilot agent you're talking about! The two types of Copilot agent that you can create Creating simple Copilot agents Creating Copilot Studio agents (chatbots or virtual agents) (this blog)

Explaining the two different types of Copilot Studio agent

Part three of a three-part series of blogs

Copilot allows you to create chat bots called "agents", but these can be either simple conversational tools or sophisticated bots depending on which type of Copilot agent you're talking about!

The two types of Copilot agent that you can create
Creating simple Copilot agents
Creating Copilot Studio agents (chatbots or virtual agents) (this blog)

Posted by Andy Brown on 02 September 2025

In this blog

Starting to create a Copilot Studio agent
Basic Copilot Studio agent properties
Creating and using topics
Four ways in which Copilot Studio agents are like Power Automate flows
The central problem

In addition to the simple agents described in the previous part of this blog, you can also create much more complex chatbots or Copilot Studio Agents (they used to be called Power Virtual Agents):

If you want to learn how to create Copilot Studio agents, consider booking yourself or your team onto our three-day Creating Copilot Studio Agents course.

Starting to create a Copilot Studio agent

To create an agent, you first need to be in Copilot Studio:

This is what you should see (here I've created a Copilot Studio agent called Ask Olly Owl to help clients choose the right training course).

Basic Copilot Studio agent properties

A Copilot Studio agent initially appears similar to a simple Copilot agent (aargh, these names!):

Here I've said what my agent is trying to achieve (helping people choose the right course) and how it will do it.

One big difference is that for a Copilot Studio agent you can restrict it to the knowledge sources you specify:

I've said that the agent can only get information from the Wise Owl website.

Creating and using topics

Any interaction that you have with a chatbot may consist of a number of different phases. In Copilot Studio these phases are called topics:

Here we're about to edit the Conversation Start topic (which is triggered when a user initiates a conversation).

As an illustration, here's the start of our topic governing what happens when a conversation begins:

The bot will introduce itself and then ask the user whether they wish to continue.

Four ways in which Copilot Studio agents are like Power Automate flows

If you already know Power Automate you'll find it easier to learn how to create Copilot Studio agents. They have at least 4 things in common. Firstly, the layout of a topic is similar to that of a flow:

This could at first glance be a Power Automate flow, but it's actually a test in a Copilot Studio agent topic.

Secondly, the expression language is the same:

Expressions like this one are written in WDL (Workflow Definition Language), which is also used for expressions in Power Automate and Power Apps.

Thirdly, you can show adaptive cards within a Copilot Studio agent (as you can in Power Automate):

You can use the built-in adaptive card designer to create miniature forms (for example, giving the user a choice of options).

And fourthly, to record choices made by users of your Copilot Studio agent you can create and call Power Automate flows:

Creating a flow within Copilot Studio agent.

The central problem

The big problem when creating an agent is to get the balance right between two opposite forces:

Approaches	The issues	Typical opening gambit
Control	Programmers (like myself) will traditionally want to control everything that a user can do, so that there are no exceptions, so we like to give choice lists rather than ask open questions.	Choose (from a fixed list of topics) which one that you would like to get training in
Relying on AI	At the other end of the scale you can delegate all control to your AI engine by asking open questions. The problem with this is that it can be hard to work out what the user's reply was.	What would you like to learn?

Suppose that you ask the second question ("What would you like to learn?"). You might then try to extract a subject from a user's answer. Here are some cases you'd have to consider:

Sample answer	What you should take from this
I already know VBA, and want to use this base knowledge to learn Python	That the user wants to learn Python (but how can your AI tools pick this out from the answer?).
I want to learn how to create web-based interactive reports	This person probably wants to learn Power BI Desktop, but how can you tell?
I want to learn how to program within Excel	How can you be sure you pick out VBA training, not Excel training?

Copilot Studio agents contain a range of tools to help you decipher a user's meaning:

An example: determining what a user meant, not what they said.

It's not easy writing bots to replace humans!

Parts of this blog
The two types of Copilot agent that you can create Creating simple Copilot agents Creating Copilot Studio agents (chatbots or virtual agents) (this blog)

Some other pages relevant to the above blogs include:

AI (Artificial Intelligence) training

This blog has 0 threads

Add a new post

ALL BLOGS

ARTIFICIAL INTELLIGENCE BLOGS

ARTIFICIAL INTELLIGENCE (AI) BLOGS

Starting to create a Copilot Studio agent

Basic Copilot Studio agent properties

Creating and using topics

Four ways in which Copilot Studio agents are like Power Automate flows

The central problem

Head office

London

Manchester