Building Copilot-style experiences in Python with the Microsoft Teams SDK

Announcement: Python developers can now build Teams-native, Copilot-style experiences using the Microsoft 365 Agents SDK, with support for Microsoft Teams as a first-class interaction surface.

That’s a meaningful shift for teams already building AI orchestration, retrieval, and automation in Python. Instead of forcing a stack change, they can bring those services into the place where users already chat, meet, approve, and act: Teams.

Why it matters:

• Build assistants that live inside Teams, not beside it
• Reuse existing Python AI/backend services
• Add conversation context, enterprise identity, and workflow actions to the experience

This is the difference between a generic web chatbot and a Teams-native assistant. The goal isn’t just “LLM + chat UI.” It’s contextual help inside real collaboration flows: incident triage, meeting follow-up, knowledge lookup, approvals, and task execution.

A simple mental model: Teams sends an activity → your Python app processes context and business logic → your app responds using Teams interaction patterns like messages, cards, and actions.

Conceptually, it looks like this:

# Minimal Teams activity endpoint: extract user/conversation context and route to AI orchestration
from fastapi import FastAPI, Request

app = FastAPI()

def orchestrate(prompt: str, user_id: str, conversation_id: str) -> str:
    return f"Hi {user_id}, I can help with: {prompt[:80]}"

@app.post("/api/messages")
async def messages(req: Request):
    activity = await req.json()
    text = activity.get("text", "")
    user_id = activity.get("from", {}).get("id", "unknown")
    conversation_id = activity.get("conversation", {}).get("id", "unknown")
    reply = orchestrate(text, user_id, conversation_id)
    return {
        "type": "message",
        "conversation": {"id": conversation_id},
        "text": reply,
    }

Important: this is only a conceptual sketch. A production Teams app requires proper activity handling, authentication, and adapter/framework plumbing rather than treating a raw JSON response as the full integration model.

For Python teams, the practical implication is clear: you can now build collaboration-native AI experiences for Microsoft 365 without abandoning the language and services you already use.

If you’re building in this space, what Teams-native use case would you ship first?

#MicrosoftTeams #Python #EnterpriseAI

Try it yourself

Run this tutorial as a Jupyter notebook: Download runbook.ipynb (15 cells, 12 KB).