AI Labs GitHub

Copilot Studio · Copilot Cowork · Copilot+ PC

Build AI agents by doing it, one lab at a time

Three hands-on tracks. Grow one agent from a few lines of instructions into a team of agents that act, ask for approval and start on their own. Hand real multi-step work to Copilot Cowork. And run AI on the PC itself, on the NPU of a Copilot+ PC, where the data never leaves the device.

Start the Copilot Studio trackStart the Cowork track

Browse the tracks

Learning path · 10 labs · about 8 hours

From simple agents to autonomous agents

One storyline, the Contoso IT Service Desk. Start in Agent Builder with instructions and company knowledge, move to Copilot Studio, then add connectors, MCP tools, workflows with human approval, skills, memory, connected agents and triggers. Finish by evaluating, publishing and governing what you built.

  1. Prepare the lab environment
  2. Build your first agent with natural language
  3. Ground your agent in company knowledge
  4. Capabilities, skills and the jump to Copilot Studio
  5. Your first Copilot Studio agent
  6. Give your agent hands: connectors and MCP tools
  7. Workflows and human approval
  8. Skills, memory and connected agents
  9. Build an autonomous agent with triggers
  10. Evaluate, publish and govern your agents

Start the Copilot Studio track ›

Workshop · 6 labs · about 3.5 hours

AI automation with Copilot Cowork

Cowork plans and carries out multi-step work across your mail, calendar and files, then hands you a finished result. Run your first tasks, organize files and meetings, then build custom skills, a personal dashboard and a plugin you can publish. Each lab has a teaching edition for trainers.

  1. Getting Started with Copilot Cowork
  2. AI-Powered Calendar Optimization and Executive Meeting Coordination
  3. Intelligent File Organizer
  4. Building Custom Cowork Skills (Read-only)
  5. Building a Personal Work Dashboard with Copilot Cowork (Read-only)
  6. Building and Publishing a Custom Copilot Cowork Plugin (Read-only)

Start the Cowork track ›

Hands-on · 11 apps · Copilot+ PC

On-device AI apps for Copilot+ PC

AI that runs on the PC, on the NPU, through the Windows AI APIs: Phi Silica, on-device OCR, image description and a local semantic index. Build and install real desktop apps for finance, fraud, HR, retail and personal knowledge, where the data never leaves the device.

Start here

  1. Architecture shared by all apps
  2. Deployment guide: from git clone to a running app

The apps

Start the Copilot+ PC track ›

How every lab works

  1. The scenarioA Contoso request: why this lab matters.
  2. The conceptsJust enough theory, in plain words, with an analogy.
  3. Step by stepReal screenshots of the current UI, field by field.
  4. Test itPrompts to try, and the trace that proves what the agent did.
  5. TakeawaysWhat to remember, and what to tell a customer.

What you need

Copilot Studio track

A test Microsoft 365 tenant, a Microsoft 365 Copilot license, Copilot Studio access with Copilot Credits, and a Power Platform environment with Dataverse.

Cowork track

A Microsoft 365 account with Copilot and access to Copilot Cowork. Sample files are provided in the repository.

Copilot+ PC track

A Copilot+ PC with Windows 11 24H2 or newer to run the AI features, plus Git and the .NET 8 SDK. No Visual Studio, no cloud subscription.

For trainers

Run these labs as a workshop

Each track has a trainer guide: agendas, what to check before you deliver, common problems and their fixes, and discussion questions.