Share via

Key concepts - Authoring agents

  • Article

Creating an agent with Copilot Studio is easier than ever. Whether you're new to conversational AI or a seasoned developer, our intelligence platform is with you and your team every step of the way.

How agent conversations work

Copilot Studio uses customized NLU model and AI capabilities to understand what a user types, then responds with the best agent topic. An agent topic is a portion of a conversational thread between a user and the agent. Topics are linked together to form nodes. For more information, see Create and edit topics.

For example, you might create an agent for your customers to ask common questions about your business. Your support overhead is reduced by deflecting support calls. In the agent, you can create a topic about your store's opening hours and name the topic Store hours.

When a customer asks a question such as "When do you open?" or "What are your opening hours?", the agent uses natural language understanding (NLU) to understand the intent behind the question. The agent matches that intent to the best topic, the Store hours topic.

The agent follows the conversation flow, which is a group of connected nodes, that you define in the Store hours topic. These questions use if/else arguments, or logic gates, to determine which store the customer wants. The final output of the topic shows the hours and contact information for that store's location.

However, you can't anticipate all the types of questions your customers ask. To help mitigate this, Copilot Studio incorporates powerful new AI-powered capabilities that use the latest advancements in NLU models. With your agent linked to a public or Bing-indexed website, your agent automatically generates responses. These responses are conversational, plain language, and don't depend on the agent builder to create topics for every eventuality.

As well, when AI general knowledge is enabled, your agent can access information not in its website or other knowledge sources.

Your agent uses AI powered by the Azure OpenAI GPT model, also used in Bing, to create agent topics from a simple description of your needs. Similarly, you can modify and update any topic in your agent by describing the changes you want to make.

Accessibility

The agent authoring canvas is built for accessibility in accordance with Microsoft accessibility guidelines and supports standard navigational patterns.

Routing anchors

For complex scenarios, users who author content with keyboard navigation can use routing anchors.

Press Enter or Space on a route anchor to go into editing mode for that route. Once in the editing mode, you can Tab to other node anchors.

With an anchor selected, you can:

  • Press Delete to delete the route. The target node becomes an orphaned node if this route is the only route leading to that node.
  • Press Escape to come out of Tab mode and resume normal tabbing.
  • Press Enter or Space on another node to reconnect the route to the new node. This node has the same source node, but a new target node.
  • Press Tab to go to the next available location to drop this route.

At any time, pressing these hotkeys on a route anchor reads out the properties of the route:

  • Alt + Shift + A reads out the source node.
  • Alt + Shift + B reads out the target node.

Note

If you're using a screen reader, the Settings navigation menu might not get read as "Collapsed", but it's an expandable control which can be selected and expanded. This is a known issue with our menu control, which will be fixed at a future date.