Learn

Foundations

What M365 Copilot, Copilot Studio, and agents actually are — how they work, what you need to use them, and where they fall short. No hype.

M365 Copilot Copilot Studio What are Agents Getting Started
1
M365 Copilot

What is Microsoft 365 Copilot?

M365 Copilot is an AI assistant embedded across your Microsoft 365 apps. It sits inside Teams, Outlook, Word, Excel, PowerPoint, and more — and it uses your organisation's data to give contextually relevant responses. Here's what that actually means in practice.

How it works

Under the hood, M365 Copilot connects a large language model (similar in capability to GPT-4) to Microsoft Graph — the data layer that sits across your tenant and knows about your emails, calendar, Teams chats, files, and meetings. When you ask Copilot a question, it retrieves relevant content from Graph and uses the LLM to generate a response.

This means it can answer questions like "What did I miss in yesterday's meeting?" or "Summarise the key decisions in the Q3 review document" — because it has real access to your data. It doesn't just work from generic knowledge; it works from your organisational context.

Where it works
TeamsSummarises meetings in real time or after the fact. Recaps what you missed, extracts action items, and can answer questions about meeting content.
OutlookDrafts emails, summarises long threads, surfaces key action items, and helps you respond faster.
WordDrafts documents from prompts, rewrites and improves existing content, summarises long documents.
ExcelAnalyses data, explains formulas, surfaces trends and anomalies, generates charts and summaries.
PowerPointCreates presentations from a brief or from an existing document. Adds speaker notes and reformats slides.
Microsoft 365 ChatA cross-app assistant that can reason across your emails, meetings, files, and Teams — a unified view of your work context.
What it does well

Meeting summarisation in Teams is genuinely useful — particularly for long recurring meetings and calls you can't attend. The quality is consistently good when the transcript is clear. Email drafting in Outlook saves time on routine correspondence. Summarisation and document generation in Word is solid, especially for first drafts. For knowledge workers in large organisations dealing with high communication volume, the time savings are real.

What it doesn't do
Honest limitations
It can't take actions on your behalf. Copilot can summarise, draft, and generate — but it can't send emails, book meetings, or update records without you confirming. That's what agents are for.

It hallucinates. Like all LLMs, it sometimes generates plausible-sounding but incorrect information. Always review outputs, especially in Excel analysis or factual document generation.

It respects your existing permissions. This is actually good news for governance — it can only surface data you already have access to. But it means it won't give everyone visibility of everything.

It requires specific licensing. You need a Microsoft 365 E3 or E5 subscription plus the Microsoft 365 Copilot add-on. As of mid-2025, the add-on is $30 USD per user per month.
What you need
Requirements
License: Microsoft 365 E3 or E5 (or Business Standard/Premium in some configurations) plus the Microsoft 365 Copilot add-on.

Admin enablement: Your IT admin needs to assign the license and enable Copilot in the admin centre. There are data access and governance settings to configure first.

Data readiness: Copilot surfaces your org's data. If SharePoint permissions are chaotic or sensitive files are over-shared, Copilot will reflect that. Data hygiene matters more than people expect.
Next step for IT admins Learn how to enable and govern M365 Copilot in your tenant — including data access controls, sensitivity labels, and usage reporting.
IT Admin pathway →
2
Copilot Studio

What is Copilot Studio?

Copilot Studio is Microsoft's low-code platform for building custom AI assistants and agents. It's part of Power Platform, and it gives you the tools to create conversational AI that connects to your own data, systems, and processes — without writing application code.

What it is

Where M365 Copilot is Microsoft's AI assistant for your Microsoft 365 apps, Copilot Studio is the platform you use to build your own AI assistants. You can create a bot that answers HR policy questions from your SharePoint intranet, a customer support agent for your website, or an internal IT helpdesk that creates tickets automatically. The scope is broad.

It uses a combination of rule-based dialogue flows (you define topics and what happens when certain phrases are detected) and generative AI powered by Azure OpenAI (the agent can answer questions it hasn't been explicitly programmed for, using your configured knowledge sources). You control how much autonomy the AI has.

Two main use cases
Standalone copilotsBuild a purpose-built AI assistant for a specific use case: HR Q&A, IT helpdesk, customer support, onboarding guide. Deploy to Teams, SharePoint, a website, or other channels.
Extending M365 CopilotAdd custom topics and plugins to M365 Copilot so it can answer questions about your specific systems, trigger your business processes, or access data not in Microsoft Graph.
Knowledge sources

You tell Copilot Studio where to look for answers. Options include:

→
SharePoint sites — index pages and documents from your SharePoint sites for Q&A
→
Uploaded files — PDFs, Word docs, and spreadsheets uploaded directly
→
Public websites — crawl external web pages (useful for product documentation or policy sites)
→
Dataverse — query structured data in your Power Platform environment
→
Custom connectors — call any API to pull in data from external systems
Taking actions

Copilot Studio integrates with Power Automate, which is where it gets its ability to do things rather than just answer questions. You can trigger flows that create calendar events, update CRM records, send notifications, create support tickets, or write to any connected system. This is what separates a knowledge-base bot from a genuine agent.

What it doesn't do
Honest limitations
Generative answers can be imprecise. When the AI generates an answer from your knowledge source, it can misread or misinterpret content. Testing and tuning the quality of responses takes real time — don't assume it'll be great out of the box.

Complex logic needs Power Automate. Copilot Studio handles conversation flow. Anything with multiple conditions, data transformations, or integration complexity needs a properly designed Power Automate flow behind it.

It doesn't train a custom model. Copilot Studio retrieves and synthesises from your configured knowledge sources. It doesn't retrain the underlying LLM on your data — there is no fine-tuning here.

Licensing is per-message or per-session. Usage is metered. For high-volume deployments you need to plan your message costs. Check current Microsoft pricing before building — it changes.
Ready to build? The developer pathway walks through creating your first Copilot Studio agent, from environment setup to Teams deployment.
Developer pathway →
3
Agents

What are agents?

"Agents" has become one of the most overloaded words in enterprise AI. Here's a grounded definition: an agent is an AI system that can take a sequence of actions to accomplish a goal — with minimal human direction at each step. That's a meaningful step up from a chatbot.

Chatbots vs agents

A chatbot responds to inputs. An agent pursues goals. The practical difference: a chatbot answers "what's the leave policy?" — an agent could receive a leave request, check the policy, look up the employee's remaining allowance, draft the approval email, and create the calendar block. Same underlying technology, very different design.

Agents have access to tools — APIs, flows, databases — and they decide which tools to use, in what order, to accomplish the task. The "decision" is made by the LLM based on the instructions you give the agent and the tools you've made available to it.

Types of agents in Microsoft's world
Copilot Studio agentsBuilt in Copilot Studio, triggered by a user message or a system event. Can take multi-step actions using Power Automate flows. Deployed to Teams, SharePoint, or web.
Autonomous agentsRun without a user initiating a conversation. Triggered by a schedule, a data change, or an external event. Can monitor, decide, and act on your behalf.
M365 Copilot agentsExtensions to M365 Copilot built in Copilot Studio or Teams Agent Builder. Give Copilot access to your specific data or the ability to trigger your processes.
Multi-agent orchestrationMultiple agents working together — one orchestrator delegates tasks to specialist agents. Microsoft is building this out with the Copilot Studio agent builder. Early stages.
What makes an agent different in practice
1
Multi-step reasoning. Instead of answering one question, an agent plans a sequence of steps and executes them in order.
2
Tool use. Agents can call APIs, run queries, trigger automation flows — they can interact with your systems, not just describe them.
3
Background operation. Agents can run autonomously — monitoring a data source, checking conditions, and taking action when needed — without a human prompting each step.
4
Context retention. Agents can hold context across a session and refer back to what's happened earlier in a task sequence.
What to watch out for
Honest limitations
"Autonomous" still requires careful design. Agents that act without human confirmation can cause real problems if the instructions are ambiguous, the data is unexpected, or the tool it calls has broad permissions. Design for failure, not just the happy path.

The action space is what you build. An agent can only do what you've connected it to. The "intelligence" is in the reasoning and planning — the agent's actual power comes from the quality of the tools you give it.

Governance matters more as autonomy increases. Know what data your agent can access, what systems it can write to, and who has oversight. This is especially important for agents that run on a schedule.

The landscape is still maturing. Microsoft's agentic AI features are evolving quickly. What's possible in Copilot Studio today will look different in 12 months. Build flexibly.
See agents in action Browse real-world use cases — HR agent, IT helpdesk agent, and more — with what you need and what realistic outcomes look like.
View use cases →
4
Getting Started

How to actually get started

Each tool has a different entry point. Here's the practical starting point for each one — what to do on day one, depending on your role and what you're trying to achieve.

Getting started with M365 Copilot
1
Confirm your licensing. Check with your IT admin that you have the M365 Copilot add-on assigned. If you don't have it, you'll need to request it or check whether a trial is available.
2
Start in Teams. The best first experience is meeting summarisation. Join or attend a recorded meeting and ask Copilot in the meeting recap to summarise key decisions and action items. It's immediately useful.
3
Try M365 Chat (Business Chat). Open Microsoft 365 Chat and ask a work question that spans multiple contexts — "What projects am I involved in this week?" It shows the cross-app reasoning capability.
4
Use it in Outlook. Open a long email thread and ask Copilot to summarise it. Then draft a reply. The friction reduction for high-volume inboxes is immediately visible.
5
Don't over-prompt on day one. Copilot works best with clear, specific prompts. "Write me a document about X" produces mediocre output. "Summarise the key points from [specific document] in 5 bullet points for a senior audience" produces much better results.
Getting started with Copilot Studio
1
Sign up for a free trial. Go to copilotstudio.microsoft.com and sign in with a Microsoft work account. You'll get a trial environment with enough capacity to build and test.
2
Create a blank agent. Give it a name, a description, and tell it its role in the instructions field. The instructions are the most important thing to get right — think of them as the agent's job description.
3
Add a knowledge source. Connect it to a SharePoint site or upload a PDF. Then test it in the built-in test panel. Ask questions that should be answerable from your content and see how it does.
4
Add a simple action. Create a Power Automate flow that does something basic — send a Teams message, create a Planner task — and wire it to the agent. Now it can do things, not just answer.
5
Publish to Teams. Once you're happy with the behaviour, publish it. Deploy to Teams as an app your colleagues can talk to.
Getting started with agents
1
Start with the Copilot Studio templates. Microsoft provides starter templates for common agent scenarios. Pick one close to your use case and modify it rather than starting from scratch.
2
Define the scope clearly. What is the agent for? What can it do? What can't it do? Write the agent instructions as if you're briefing a new employee — specific, bounded, and clear about edge cases.
3
Build the tools first. The agent's capability is in the actions available to it. Identify the 2–3 things you want it to be able to do and build those Power Automate flows before worrying about conversational quality.
4
Test adversarially. Give the agent inputs it shouldn't be able to handle and confirm it fails gracefully. What happens when it doesn't understand? What happens when the data it needs isn't there?
5
Add a human escalation path. For any agent handling real business processes, define the point at which a human takes over. Don't automate without a safety valve.
Your next step
Not sure which tool to focus on first? The Pathways section has structured learning tracks matched to your role — IT admin, developer, or business user.
Stay in the loop

Get new articles
when they're published.

Practical content about Microsoft AI as the platform evolves. No noise.