Two announcements at the end of September changed how I talk about Copilot with customers. On 24 September 2025 Microsoft expanded model choice in Microsoft 365 Copilot, including Anthropic models. Five days later, on 29 September, it introduced Agent Mode and Office Agent.
1️⃣ Model choice, including Claude
Copilot is no longer tied to one AI provider. Anthropic’s Claude models became available alongside OpenAI’s models, starting with the Researcher agent and Copilot Studio.
- You can pick the model that works best for a specific task.
- Copilot Studio makers can build agents on the model that fits their scenario.
- Admins decide whether Anthropic models are enabled in the tenant.
My advice to IT: review the admin setting and your data processing requirements before enabling it, and communicate the decision clearly.
2️⃣ Agent Mode: from assistant to co-worker
Agent Mode in Excel and Word lets Copilot work inside the document: planning steps, building the spreadsheet or the draft, checking its own work and letting you steer along the way. Microsoft calls this “vibe working”: you describe what you want and work together with Copilot to get there.
3️⃣ Office Agent in Copilot chat
Office Agent lets you create complete PowerPoint presentations and Word documents from a conversation in Copilot chat, with research and structure included. It is one of the first places where you can see Anthropic’s models working inside Microsoft 365.
What I recommend
- Decide your multi-model policy now, before users ask for it.
- Let a few power users try Agent Mode in Excel and Word through the Frontier program.
- Collect real examples of documents and analyses that Agent Mode improved.
My take
Put together, these two announcements move Copilot from “an assistant that helps” to “a co-worker that does”. The organisations that learn how to delegate work well to agents will pull ahead.
Would you enable Anthropic models in your tenant? Why or why not? 🤔
Leave a Reply