∧

What is an agent?

Agent flows offer a powerful way to automate repetitive tasks and integrate your apps and services. Agent flows can be triggered manually, by other automated events or agents, or based on a schedule.

With Copilot Studio, you can create agent flows by using natural language or a visual editor.

You can run agent flows as standalone automations. You can also configure an agent flow to trigger from an agent as a tool, and return results to the same agent.

Copilot Studio agents use customized NLU model and AI capabilities to understand what a user types or says, then respond with the best topic. A topic is a portion of a conversational thread between a user and the agent.

For example, you might create an agent for your customers to ask common questions about your business. Your agent reduces support overhead 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. Some nodes can ask questions, while others use conditions (if/else) to determine which store the customer wants. The final output of the topic shows the hours and contact information for that specific store.

However, you can’t anticipate all the types of questions your customers ask. To help mitigate this issue, Copilot Studio incorporates powerful AI-powered capabilities that use the latest advancements in NLU models. Once your agent is linked to knowledge sources, it can automatically generate responses. These responses are conversational, plain language, and you don’t need to create topics for every eventuality.

You can also choose to let your agent access information outside its knowledge sources.

Copilot Studio can use AI powered by the Azure OpenAI GPT model, also used in Bing, to create 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.

• Prompt-and-response agents
  Prompt-and-response agents are AI tools designed to perform specific tasks based on a user’s input or “prompt.” These agents process the given input and generate a corresponding response, facilitating a back-and-forth exchange with the user. They can be used in various contexts, such as chatbots, virtual assistants, or specialized AI systems for business applications.

Prompt-and-response agents operate based on the type of data they have access to, as well as the set of predefined rules that determine their behavior. They can react quickly to any changes in the data, their rules, or the context in which they operate.

Because they are meant to simulate natural conversation, they are commonly used in scenarios that require immediate responses to user inquiries or requests, such as customer service, for instance. Prompt-and-response agents are effective for organizations looking to streamline routine interactions and enhance customer support—without adding more complexity to their existing systems.

• Task agents
  Cognitive agents are designed to mimic human thought processes. They use machine learning and natural language processing to understand, learn, and adapt to user behavior over time, which makes them useful for analyzing historical data and taking action based on their learnings.

Cognitive agents are used in applications such as virtual assistants, like Siri and Alexa, that can learn from user preferences and continuously improve their responses. In businesses, cognitive agents can analyze customer data to deliver tailored recommendations that support more informed decision-making. Using data-driven insights, these agents help organizations create more personalized user experiences, improving customer satisfaction and engagement.

• Autonomous agents
  Autonomous agents have their own distinct roles and capabilities. While they operate independently as an entity, they’re also able to interact and collaborate with other agents to solve complex problems, optimize larger processes, or achieve a specific goal.

These multi-agent systems are frequently used in logistics and supply chain management, where autonomous agents can dynamically orchestrate tasks such as inventory management, shipment tracking, and resource allocation. They can also be applied to large-scale environments such as smart cities, where multiple agents manage traffic flow, public transportation, and energy distribution while also learning from the data. In the business world, organizations can use autonomous agents to enhance overall efficiency across departments.