Pathways

Learning tracks

Structured, role-matched paths through the Microsoft AI landscape. Know exactly where to start and what to focus on for your situation.

IT Admin Developer Business User
IT Admin Developer Business User
IT Admin

Enable and govern Microsoft AI in your organisation

You're responsible for tenant health, licensing, data governance, and security. Your job with Microsoft AI isn't to build agents — it's to make sure Copilot is enabled safely, the data estate is in good shape, and the organisation can use these tools without creating compliance risk.

Steps 5
Focus M365 Copilot, Tenant governance, Copilot Studio policies
Tools needed Microsoft 365 Admin Centre, Purview, Entra ID

This pathway assumes you have global admin or equivalent access to your Microsoft 365 tenant and familiarity with the admin centre.

1
Understand what Copilot can see in your tenant
Before enabling Copilot for anyone, audit what data is accessible to who. Copilot surfaces content users already have permission to access — which means overshared SharePoint sites, broad security groups, and poorly labelled documents will all become visible through Copilot queries. Run a SharePoint access review first.
Key actions: SharePoint permission audit, check site-wide sharing links, review external sharing settings, identify broadly accessible sensitive content.
M365 AdminPurviewSharePoint Admin
2
Configure Microsoft Purview sensitivity labels and DLP
Copilot respects sensitivity labels — content marked as Confidential won't be summarised or surfaced in responses to users who don't have access. Set up and enforce sensitivity labels before rolling out Copilot broadly. Data Loss Prevention policies should also be configured to catch Copilot-specific scenarios.
Key actions: Define label taxonomy, apply auto-labelling policies, configure Copilot-specific DLP policies in Purview, test label enforcement with Copilot responses.
Microsoft PurviewDLPSensitivity Labels
3
Enable M365 Copilot and assign licenses
Once data governance is in reasonable shape, enable the Copilot add-on in the Microsoft 365 admin centre and assign licenses. Start with a pilot group — ideally a cross-functional set of power users who will give you useful feedback and can champion adoption. Don't roll out to everyone at once.
Key actions: Purchase and assign Copilot licenses, enable Copilot in M365 admin, set up pilot group, configure usage policies (what Copilot can and can't do), communicate to pilot users.
M365 AdminLicensing
4
Set up Copilot usage reporting and monitoring
Microsoft provides Copilot usage analytics in the Microsoft 365 admin centre and through Viva Insights. Set up dashboards to track adoption, feature usage by app, and user activity. This data is essential for ROI measurement and for identifying where adoption support is needed.
Key actions: Enable Copilot usage reports in M365 admin, configure Viva Insights dashboard, set up retention policies for Copilot interaction logs, define what usage signals you'll track for success.
Viva InsightsM365 AdminReporting
5
Govern Copilot Studio deployments
If developers or power users are building agents in Copilot Studio, you need governance guardrails. This means Power Platform environment policies, data loss prevention for connectors, and approval processes for publishing agents to Teams or SharePoint. Copilot Studio agents with broad permissions can cause real damage if not governed properly.
Key actions: Define Power Platform DLP policies, create environment strategy (dev / test / prod), set Copilot Studio publishing policies, require security review for agents that write to business systems.
Power PlatformCopilot StudioDLP
Want the foundational knowledge first? The Learn section covers what M365 Copilot is and how it works — useful context before diving into admin configuration.
Read the explainer →
Developer

Build agents and extend Microsoft AI with Copilot Studio

You're building solutions — whether that's as a Power Platform developer, a consultant implementing for clients, or a software developer extending into the Microsoft AI space. Your focus is on what's buildable, how it integrates with existing systems, and what the resulting solution actually does reliably.

Steps 5
Focus Copilot Studio, Power Automate, agent design
Prerequisites Power Platform basics helpful but not required

This pathway works whether you're new to Copilot Studio or transitioning from the classic Power Virtual Agents.

1
Set up your development environment
You need a Power Platform environment to build in. If you don't have a developer tenant, create one at developer.microsoft.com/en-us/microsoft-365/dev-program. Set up a dedicated development environment in Power Platform admin — don't build in your default environment or in production.
Key actions: Create Microsoft 365 developer tenant (or use existing), provision a Power Platform dev environment, enable Dataverse, sign in to copilotstudio.microsoft.com.
Power PlatformCopilot Studio
2
Build a knowledge-based agent from scratch
Create a new agent in Copilot Studio. Write clear, specific agent instructions — this is the most impactful thing you'll do. Add a knowledge source (a SharePoint site or uploaded document) and test it thoroughly in the test panel. Learn what "generative answers" mode does vs. classic topics.
Key actions: Create blank agent, write system instructions, add SharePoint knowledge source, test with 20+ varied questions, review which questions it handles vs fails, refine instructions based on failure patterns.
Copilot StudioGenerative AIKnowledge Sources
3
Add Power Automate actions
Connect your agent to Power Automate to give it the ability to take actions. Start with something simple: create a task in Planner or send a Teams notification. Then progress to something that reads from a data source — a SharePoint list, Dataverse table, or an external API via a connector. This is where agents become genuinely useful.
Key actions: Build a Power Automate cloud flow triggered from Copilot Studio, pass user input as dynamic parameters, handle success and error responses, surface the result back in the conversation.
Power AutomateConnectorsActions
4
Design for reliability and failure
Good agents don't just handle the happy path. Build explicit fallback topics for when the AI doesn't understand. Add graceful error handling when Power Automate flows fail. Define escalation paths — how does a user reach a human when the agent can't help? Testing adversarially is just as important as testing the intended flow.
Key actions: Create fallback topic, test edge cases and off-topic inputs, add flow error handling with user-friendly messages, implement human escalation, load test with volume if deploying broadly.
Error HandlingTestingUX Design
5
Deploy and extend to Teams or M365 Copilot
Publish your agent. For internal tools, Teams is usually the right channel — users are already there and adoption friction is low. If you want to extend M365 Copilot with your agent's capabilities, publish it as an M365 Copilot plugin. This lets M365 Copilot invoke your agent's actions when relevant to a user's query.
Key actions: Publish to Teams via app package, configure channel-specific settings, submit to Teams admin for approval, optionally configure as M365 Copilot plugin, set up usage analytics.
TeamsM365 CopilotDeployment
See what you can build The Use Cases section has detailed scenarios for agents — HR onboarding, IT helpdesk, and more — including what you need and what realistic outcomes look like.
Browse use cases →
Business User

Get real value from M365 Copilot in your daily work

You're not building agents — you're a professional who works in Microsoft 365 every day and wants to understand whether Copilot is worth the hype, and how to actually get useful outputs from it. This track is about building practical prompting skills and identifying the use cases where Copilot genuinely saves you time.

Steps 5
Focus M365 Copilot in Teams, Outlook, Word, Excel
Prerequisites M365 Copilot license from your IT admin

If you don't have a Copilot license, ask your IT admin. Some organisations are running pilots — you may be able to get access for evaluation.

1
Start with meeting summaries in Teams
The quickest win with Copilot. After any transcribed Teams meeting, open the recap tab and ask Copilot to summarise key decisions and action items. Compare what it produces against what you'd have written manually. This single use case alone justifies the learning investment for most people who sit in a lot of meetings.
You need: Teams meetings with transcription enabled (your admin may need to turn this on). Ask IT if you're not sure.
TeamsMeetings
2
Use Copilot in Outlook to manage your inbox
Open a long email thread and ask Copilot to summarise it — "What are the key points and any open questions in this thread?" Then try drafting a reply: give Copilot context about what you want to say and let it draft. You'll spend your time editing rather than writing from scratch. Works well for routine, professional correspondence.
Tip: The more specific your prompt, the better the output. "Draft a polite reply declining this meeting request, mentioning I'll follow up next week" produces much better results than "write a reply".
OutlookEmail
3
Learn to write better prompts
Copilot's output quality is directly tied to how you prompt it. The shift in mindset is this: you're not asking a search engine, you're briefing a capable but literal assistant. Good prompts include context, the intended audience, the format you want, and any constraints. Bad prompts are one-liners with no context.
Framework to try: Goal → Context → Constraints → Format. "Summarise [X] (goal) in plain English for a non-technical audience (context), in under 200 words (constraint), as bullet points (format)."
PromptingSkill-building
4
Use Copilot in Word and Excel for documents and data
In Word, try asking Copilot to draft a first version of a document — a project summary, a stakeholder update, a policy brief. You'll get a reasonable first draft to edit rather than a blank page. In Excel, highlight a data range and ask Copilot to identify trends or explain what the data shows. Useful for tables you've been handed but haven't dug into yet.
Be cautious with Excel analysis: always verify Copilot's interpretations against the raw data. It can misread complex multi-sheet workbooks or draw spurious correlations.
WordExcelDocuments
5
Try Microsoft 365 Chat for cross-app questions
Microsoft 365 Chat (also called Business Chat) is a Copilot interface that works across your emails, meetings, chats, and files at once. You can ask things like "Catch me up on what's happened in the Contoso project this week" or "What were the open actions from last Thursday's leadership meeting?" This is the most powerful use of Copilot for knowledge workers — when you've been away or need a cross-context briefing.
Access it at microsoft365.com/chat or within Teams as a chat thread. Note: it only works with data you already have access to — it won't reveal content you're not permitted to see.
M365 ChatCross-app
See real use cases Browse practical scenarios with realistic outcomes — meeting summaries, document generation, and more.
View use cases →
Stay in the loop

New articles and pathways
as the platform evolves.

Practical content about Microsoft AI. No noise, no spam.