Microsoft’s Copilot agents are a major leap forward in how organisations work with data and streamline business processes.

In order to get the most out of Copilot agents, it’s important to know what they are, how they work, and where they can deliver the most value.

Copilot agents are AI assistants that can be embedded across Microsoft 365 apps and platforms and can carry out multi-step tasks without user input or oversight.

They work by drawing on internal or external data sources, understanding prompts, and delivering contextual responses or performing automated actions.

These agents can be built with varying levels of technical complexity, from no-code tools accessible to everyday users to pro-code solutions that integrate with advanced external systems.

Copilot Studio agents – Built using Copilot Studio, part of Microsoft Power Platform, offering integration with external systems and full workflow capabilities. These agents are available across Microsoft 365 and require one of the following: a Copilot licence, tenant enrolment for pay-as-you-go licensing, or the purchase of pre-paid message packs.

SharePoint agents – Optimised for interrogating and responding to SharePoint content, ideal for knowledge and document-based queries. These agents are available in each SharePoint site and require a Copilot licence or tenant enrolment for pay-as-you-go licensing for SharePoint Agents.

Out of the box, Microsoft 365 Copilot is powerful, but every organisation has unique needs, systems, and ways of working. That’s where Copilot extensibility comes in.

Copilot extensibility simply means you can customise and enhance Copilot to fit your organisation. Whether it’s tailoring how it responds, connecting it to external systems, or building entirely new agents, Microsoft provides the tools to make Copilot work the way you need it to.

1. Declarative agents – Allow you to configure Copilot for specific scenarios with custom instructions, additional knowledge, and actions to automate business processes. Microsoft 365 provides the AI model & orchestration 
2. Custom engine agents – Ideal for advanced scenarios that require complex workflows, advanced orchestration, or specialised language models  
3. Copilot connectors (formerly known as Graph connectors) – Enable organisations to bring in external data allowing Copilot and agents to access and reason over a broader range of enterprise content. 

Although Microsoft 365 Copilot and Copilot agents share similarities, there are some key differences, particularly around automation, reasoning, and memory. 

While Microsoft 365 Copilot can remember preferences and context within an app, Copilot agents allow for explicit memory management, enabling more complex, guided conversations and persistent workflows across sessions.

The scenario: An employee is feeling stressed and visits the Wellbeing Knowledge Hub for advice. There is a lot of information available, and the employee needs to be directed to specific pages relating to his feelings.

Solution: A no-code Copilot Studio agent can be built to help employees find mental health support.

By referencing a curated FAQ document, the agent can respond to prompts like “I feel stressed” with compassionate advice and practical steps.

The agent may include an escalation path and could be, for example, configured to send a confidential email to the designated Mental Health First Aider on behalf of the employee if requested.

Problem: Searching for information about a company might require trawling through the company website manually, which is tedious and time-consuming.

Solution: An agent trained solely on the organisation’s public website can answer questions about case studies, services, and client sectors. This can be particularly useful in cases such as education, where a school or university’s website may contain a large amount of information.

This agent can be used internally by sales teams or externally* by prospective clients. All the instructions and formatting can be handled via prompt-based configuration.

*An agent can be made available externally, but only if you choose to publish it. By default, it’s private. This is typically done for website chatbots or client portal agents.

Problem: Responding to Requests for Proposals (RFPs) is often repetitive and time-consuming. Teams can struggle to structure their responses or risk missing key client requirements altogether.

Solution: A SharePoint Agent can analyse an RFP document and generate a structured response template, helping users get started quickly with the right tone and format. The agent can also review a completed proposal against the original RFP to ensure all points have been addressed, thereby reducing errors and saving time.

Problem: Business teams often need data insights but lack the time or skills to perform analysis using traditional tools.

Solution: The Analyst agent, one of Microsoft’s new reasoning-capable agents, can interpret complex datasets, run step-by-step logic, and generate charts or summaries in response to natural language queries.

This empowers users to extract meaning from data without relying on technical teams.

Starting in September 2025, agent usage is now measured in “Copilot Credits” instead of “messages.”
Copilot Credits have become the standard unit for tracking how agents are used.

The licensing model for Copilot agents varies:

• Microsoft 365 Copilot users receive interactive access to Copilot Studio and SharePoint agents as part of their $30/user/month licence.
• Copilot Studio pre-paid Copilot Credits subscription allows organisations to pre-purchase packs of 25,000 Copilot Credits per month
  • $200 USD per pack (equivalent to $0.008 per Copilot Credit)
  • Unused Credits do not carry over to the next month.
• Copilot Studio Pay-as-you-go (PAYG) allows organisations to pay for the actual number of Copilot Credits used during the month.
  • $0.01 per Copilot Credit
  • You can mix both models. Pre-paid packs are used first, and then PAYG kicks in.

So how much does an agent interaction cost? That depends on what the agent is doing. The number of credits consumed during a user interaction depends on several factors, including the agent’s underlying design (i.e., the processes it runs to generate responses), how frequently users engage with it, and the specific features they access. The table below outlines key differences between the available subscription models for Copilot Studio event costs.

It’s worth noting that a single interaction with an agent does not necessarily mean it is classified as a single Copilot Credit, as it’s based more on actions. If you have a multi-step automated process within an agent, and one that perhaps interacts with external data, a single query can result in numerous “Credits” being consumed in the processing of this action.

At Intelogy, we’re ready to partner with you on your AI journey, specifically with Copilot agents. We’ll start by helping you define, refine, and implement a robust AI strategy. This includes:

• Identifying use cases to find the best opportunities for Copilot agents.
• Creating quick wins that generate excitement and drive adoption throughout your organisation.

Crucially, we’ll ensure you have the necessary foundations in place. A successful launch depends on sound governance and a solid architecture.

We can help you: