Claude Opus 4.7 in Microsoft 365: Why the Office Suite Just Became the AI Battleground

The interesting story in legal AI this month is not the model. It is where the model now lives.

What Anthropic Actually Shipped

Two things landed in a three-week window. On April 16, Anthropic released Claude Opus 4.7. Pricing held at $5 per million input tokens and $25 per million output tokens. The bigger move came on May 7, when Claude went live as a native add-in across Excel, PowerPoint, and Word at general availability, with Outlook in public beta. The four add-ins share one conversation thread per user. An assumption set in a financial model carries over to the deck and to the cover memo without re-prompting.

Then on May 13, Ramp published its monthly AI Index. Anthropic hit 34.4 percent of business AI adoption in April. OpenAI dropped to 32.3 percent. A year ago the same dataset had Anthropic at under 8 percent.

The model is good. The pricing is the same. The real product is that Claude now sits inside the four apps lawyers already had open this morning.

The Microsoft 365 Thesis, and Why It Just Got Real

Here is the thesis: in any professional services firm, the workflow runs on Microsoft 365. Email is the front door. Word is where documents get drafted. Excel holds the numbers. Outlook holds the approvals. SharePoint holds the closed files. Any AI product that does not connect to that stack is a demo, not a tool.

Vendors have been saying this for two years. What changed in May 2026 is that Anthropic, the company most enterprises now trust on model quality, ratified the architecture by shipping into it directly rather than building a competing surface. Claude.ai still exists. The bet is that most paying work now happens inside Word and Excel, not in a chat tab.

That bet is consistent with the Ramp data. Anthropic's enterprise lead is coming from firms wiring Claude into the systems they already pay Microsoft for. The 34 percent number measures spend on corporate cards. People putting AI charges on the company card are doing the company's work.

A second-order effect: smaller AI vendors who built standalone web apps now have a choice. Ship a Microsoft 365 integration, or accept that their tool will live one tab away from the work, which is to say, mostly closed.

What This Changes for Legal Drafting

For a Quebec law firm, three things matter.

First, the intake problem partly solves itself. Outlook is where client requests, broker emails, and referring-lawyer threads land. A Claude agent that reads the inbox, drafts a reply, opens the attached PDF in Word, and pulls a precedent from SharePoint is a different unit of work from a chat you copy-paste from.

Second, document drafting moves from "ask the AI in a separate tab" to "ask the AI about this clause, in this document, with these tracked changes." Word has had Copilot for two years. Adding Claude as a peer model inside the same panel is what pushes the workflow over the line for firms that found Copilot's drafting quality thin. Anthropic's strength on long-context reasoning matters more for a fifty-page asset purchase agreement than for a marketing email.

Third, the audit trail gets simpler. When the model edits a Word document, the firm has a record. When it drafts a reply in Outlook, the firm has a record. The model still hallucinates and still requires lawyer review, but the artefacts of its work now sit in the same governed systems the firm already retains for seven years.

The risk that does not go away: Quebec privacy obligations under An Act respecting the protection of personal information in the private sector still apply. The data lives in Microsoft 365, but inference happens at Anthropic. Firms need a data processing addendum that covers both and a written client policy on AI-assisted drafting before they turn anyone loose.

Cleardeal as a Working Example

Cleardeal is a multi-tenant title-review application I built for Quebec legal teams. It runs on the same architectural thesis Anthropic just validated. Cleardeal connects to a firm's Microsoft 365 inbox through the Graph API, detects incoming title-review requests, ingests the attached PDFs, runs OCR and vision extraction on the encumbrances, and produces a draft DOCX letter that the lawyer reviews and sends back through the same Graph connection.

The design choice that mattered was the connection point. Cleardeal does not ask a firm to forward emails, paste threads, or maintain a separate inbox. It treats Microsoft 365 as the system of record and slots in as a workflow layer. Two years ago that was a thesis with risk attached. Microsoft might have closed the API. Anthropic might have built a competing surface. Neither happened. Microsoft kept Graph open and broadly compatible. Anthropic chose to ship into Microsoft rather than against it.

That gives products like Cleardeal a wider lane. The platform stays where the work already happens. The model gets better and cheaper. The integration is the asset.

What to Watch Next

Three open questions for the rest of 2026.

How does pricing hold up under enterprise load? Token-based billing scales with volume, and a busy M&A practice can ring up large bills if drafting volumes are not metered. Ramp's own report flags this as the central threat to Anthropic's lead.

Does Microsoft route paid Copilot users toward Claude inside Office, or quietly toward OpenAI's models in Foundry? The Microsoft and OpenAI relationship still matters here, and Microsoft has commercial reasons to keep both in play.

Does Google Workspace get an equivalent third-party agent story, or stay a Gemini-only surface? If the answer is the second, Workspace becomes a structurally weaker platform for legal tech.

The model layer commoditizes. The integration layer does not. That is the practical takeaway for anyone building or buying legal AI in 2026.