Microsoft Copilot (Microsoft) - capabilities
The safe, trusted way to bring AI into a Microsoft shop: it doesn't build its own frontier models (it runs OpenAI and increasingly Anthropic), so it sits a step further from the cutting edge - but your data stays inside Microsoft and the AI lives right where people already work. Our view from 500+ client engagements; capabilities evolve quickly.
Best at
- Living inside the M365 stack (Outlook, Teams, Word, Excel, SharePoint)
- Being the "trusted" option for Microsoft-only enterprises where data must stay in
- Automating business processes via Copilot Studio / Azure Foundry
- Office data grounding and connectors (Power BI, internal systems)
- Riding others' models - now combining OpenAI and Claude in parallel
Capability snapshot
| Capability | Verdict | What that means |
|---|---|---|
| Integration / grounding (M365) | ✅ Leads | Its true edge: AI sits where people already work, grounded in the company's own Microsoft data. |
| Data security / trust | ✅ Strong | Chosen because data stays inside Microsoft - the easy procurement story for Microsoft-only firms. |
| AI agents (Copilot Studio) | ✅ Strong | Copilot Studio combines AI with deterministic if-then flows to automate real business processes. |
| Connectors / MCP | ✅ Strong | Connects easily to internal systems like Power BI - a natural fit for grounded enterprise data. |
| Office docs (Excel / PowerPoint / Word) | 🟡 Capable | The Excel plug-in is genuinely good now; PowerPoint is noticeably weaker than Claude. |
| Reasoning / model quality | 🟡 Capable | Runs OpenAI and Anthropic rather than its own models, so it trails the labs closest to the frontier. |
| Custom assistants / agents | 🟡 Capable | Self-built agents from a single prompt up to pro-code, shareable across the enterprise. |
| Deep research | 🟡 Capable | A "model Council" runs OpenAI and Claude side by side - but gated behind the Frontier subscription. |
| Code & data analysis | 🟡 Capable | Fine for in-Office data work; for heavy Excel modelling we still reach for Claude. |
| Note-taking / meeting capture | ⚠️ Weak | Teams note-taking is restricted and inconsistent, and lags dedicated transcription tools. |
In Wouter's words
Inside PowerPoint, I find Copilot much, much less strong than Claude.
It allows OpenAI and Claude to be run in parallel, and then basically the best of both worlds are combined inside Copilot.
Watch-outs
- It doesn't make its own frontier models - it rides OpenAI and Anthropic, so it sits a step behind.
- The genuinely impressive new pieces (Cowork, Researcher, model Council) are gated behind the Frontier subscription, so most large corporates don't have them yet.
- Excel got materially better largely because it now embeds Claude - credit the model, not Copilot itself.
- Note-taking in Teams is limited and often blocked by client policy; weaker than dedicated tools.
Our take
Copilot's strength was never the model - it's the trust and the placement: your data stays inside Microsoft and the AI shows up exactly where people already work, in Outlook, Teams and Office. Because it runs other labs' models it sits a step behind the frontier, and for several tasks (PowerPoint, heavy Excel modelling) we'd still reach for Claude - though the gap is closing fast precisely because Microsoft is now embedding Anthropic and OpenAI inside Cowork, Researcher and the Excel plug-in. For a Microsoft-only enterprise that wants safe, governed AI in the flow of work, Copilot plus Copilot Studio is a serious platform; for raw capability on a single hard task, it's usually not the sharpest tool in the room.
Copilot's strength was never the model - it's that your data stays inside Microsoft and the AI shows up where people already work.
- Wouter van Haaften, WAIMAKERS
Compared head-to-head