It’s incredible how two letters – ‘AI’ – represent a wave of excitement that has captured the world’s imagination and is driving innovation among businesses. However, the term AI and several other terms used in this field can be confusing. Understanding these terms can help in making more informed decisions when using AI tools, such as Microsoft Copilot Agents.
Understanding AI Terms
Before we explain more about Copilot Agents here’s some key AI terms that will be helpful for you to understand.
Natural Language Processing (NLP) enables computers to understand human language in a meaningful way. Copilot agents use NLP to interpret the language of user queries and gather information.
Generative AI models (gen AI Models) are designed to create content based on the data they have been trained on. Large Language Models (LLMs) are a type of gen AI Model that are trained on understanding language and focused on text generation.
Copilot Agents use LLMs which are a type of NLP. LLMs are also used in the creation of Copilot Agents through the use of prompts. The agents themselves also use LLMs to understand the context of a user query and be able to generate a response.
Below is a diagram to highlight the relationships between these different terms:
There has been a lot of hype surrounding the terms ‘AI Agent’ and ‘Agentic behaviour’ both from Microsoft and the wider industry. An AI agent has the autonomy to make decisions. It can act on those decisions to reach the specific goal that has been given without requiring step-by-step instructions. It should be able to reason, act and access memory to achieve its goal without needing human input. At their core, an LLM is the brain of an AI agent and powers the system.
Microsoft & AI
Microsoft have described their specific implementation of Agents and the different types available to help businesses along their AI journey. The terminology of Copilot offerings have recently changed. Previously, ‘Copilots’ were created from within Copilot Studio now ‘Agents’ are created.
Copilot Agents are well-defined workflows that uses scoped knowledge sources. They are tailored to meet the needs of the user.
A range of Copilot Agents can be created from within Copilot Studio. From simple to advanced and have the ability to be autonomous.
It is important to keep in mind Microsoft’s definition of an autonomous agent is to ‘run in the background’. It is automatically triggered by pre-determined events as part of an automated cloud flow, for example ‘when an email arrives’.
When creating a Copilot Agent it requires specific instructions to be able to work well. The actions that it uses as part of its workflow need to be referenced within the instructions of the Agent.
Three types of Copilot Agents
The capabilities of Copilot Agents fall into three categories. Understanding the features of each category can guide you in integrating AI and using Copilot Agents to create effective business solutions.
Retrieval Agents aim to answer user questions by gathering information from connected data sources. They reason and summarise the information to be able to answer the user’s questions. For example, a Company Policy Agent that answers employees’ questions such as ‘What is the process for claiming travel expenses?’
Task Agents aim to replace repetitive tasks for users by automating workflows by taking actions when prompted to do so by users. For example, an Equipment Request Agent that assists employees in submitting and tracking a request for approval for new equipment such as headsets, whiteboards.
Autonomous Agents aim to perform tasks in the background for users. They use automatic triggers as the agent is prompted by a pre-determined event to perform tasks in the background for the user. For example, a Lead Generation Agent that has identified several new leads based on pre-determined criteria and are ready to review by the user.
Types of agents aren’t isolated, and the capabilities of agents can overlap
Retrieval and Task Copilot Agents are often combined to be able to offer the capabilities provided in both categories. For example, an Annual Leave Agent could be created to provide answers to employees such as ‘How many days of annual leave do I have remaining?’ and use the company’s data to provide this answer. It could also be configured to act on annual leave requests. For example, an employee could use the agent to request annual leave and support the associated approvals process.
Autonomous Copilot Agents are built upon Task Agents. The only difference between them is the prompt that triggers the Agent to act. For Task Copilot Agents the prompt is from the user interacting with the chat interface. Whereas for Autonomous Copilot Agents the prompt is an automatic trigger that was previously selected by the creator of the flow and can therefore run in the background.
Paving the way for more effective business solutions
By understanding key AI terms, businesses can better harness the power of AI to drive innovation and efficiency. Embracing these technologies with a clear understanding will pave the way for more effective business solutions.
It’s key to remember that Copilot Agents are knowledge workflows that are modified to meet the needs of the users.
Get in touch and have a chat with one of our AI specialists to learn more about Copilot Agents and how they can be used to transform your business.
Written by Sophie Irwin, App Maker
