What Microsoft Just Put in Word
Microsoft shipped Legal Agent for Word this week, in preview, and for a lot of boutiques the climb just got more realistic. And it’s not that Microsoft’s Agent is going to be the absolute best on market for this type of work. Well, probably not. But there’s an argument to be made that it is good enough AND it ships inside the M365 tenant the firm is already paying for, no procurement cycle, no vendor due diligence, no training on a new UI. That kind of zero friction shipment changes what every specialized vendor on the partner’s shortlist now has to clear to win the spot.
What Microsoft released last week is an AI driven contract review and redlining agent running inside Word for Windows on M365 Copilot, with a deterministic resolution layer over the edits that understands Word’s document structure - formatting, lists, tables, tracked changes. The firm’s existing M365 security setup carries through, and the agent reviews clause by clause against the firm’s own playbook, which means firms with a real playbook get the most out of it, and firms without one get Copilot with extra steps. The Robin AI hires Microsoft made earlier this year are showing up in the design choices they are making.
A lot of the conversation around this has been about which vendors survive. Now, that’s a real question, but it isn’t the one most managing partners at boutiques are concerned about. The question on their mind is what does this mean for the contract review tool decision they were about to make.
The AI Climb Got Even More Feasible
The Boutique AI Climb I wrote last week is resting on one principle, which is to squeeze what you have before buying any new AI tools. Most boutiques have more leverage in the M365 tenant they already pay for than they realize, and the climb works because the next move is almost always cheaper than the partner thinks it is. What Microsoft did this week is hand boutiques more to squeeze. Inside the same M365 tenant, on the same Copilot license, in the same Word desktop the lawyers already write in - a legal aware contract review and redlining capability that wasn’t there two weeks ago.
For boutiques doing transactional, IP, commercial work (or similar high volume work), this lands in a specific slice of matter management. Contract review and redlining is the slice where most of the buy conversations have been pointed for the past year, because the off-the-shelf vendors got noticeably good and the lawyers felt the pull. A managing partner sitting with a Spellbook or Legora quote on the desk was, until this week, weighing whether the specialized tool was worth the procurement cycle, the security review, the contract negotiation, and the months of running two tools in parallel. That weighing now has a third option on the table, which is to push what’s already in the M365 tenant to its ceiling first, see what’s left, and only then decide whether the specialized vendor still wins.
That’s not the same conversation as a month ago. The diagnosis underneath the buy decision didn’t change. The cheapest move on this one slice quietly got cheaper, and the climb the firm was already on just got a step shorter.
Good Enough: Where the Work Already Lives
The partner sitting with this decision is weighing more than which tool reviews contracts better. There are specialized vendors on the market today (Spellbook, Legora, Robin AI - whose engineers Microsoft hired in January this year), and they will keep being better on the legal-specific pieces for a while. But the buy decision at a boutique runs on more than which tool wins on the work itself. It runs on whether the better tool is “enough better” to clear the friction of bringing a separate vendor into the firm - procurement cycle, contract negotiation, security review, partner, associate, and paralegal training, the months where lawyers are running both tools in parallel because they don’t trust the new one yet, and the data sitting in two places that have to be reconciled.
I don’t think Microsoft’s tool is trying to compete with the specialist on quality of the work itself ONLY - that’s the vector the specialist still owns, especially on the matters where the quality matters most. Microsoft is competing on being good enough at the work and inside the place the work already lives, with no friction to roll out. For a lot of contract review at boutiques, that math now favors stitching what you have. The specialized vendor needs to clear a higher bar than it did three weeks ago. “Enough better” at the work to justify everything that comes with bringing it in.
The work that carries real risk is where the specialized vendor still earns its spot. Complex M&A, high-stakes commercial transactions, the redlines where one missed clause changes the deal, that’s where the boutiques will keep buying the specialist, and they should. The Microsoft tool isn’t trying to win all of that work. It’s trying to take the slice underneath it, the routine work that can be standardized (for example reviews that used to chew up paralegal capacity) and that’s the slice where a lot of boutiques are about to find a different answer than they had on the table a month ago.
Let’s Talk Practicalities
Now how can you use it? There’s a caveat because it’s US-only today. It’s also only on Windows desktop currently, so Mac, web, UK, EU, Australia not in scope yet. It’s also a part of Microsoft Foundry, so it’s a preview, not GA, so features may evolve and there’s no ship date for the full release. Firms whose contract work has already moved off Word (heavy CLM users, firms running through Clio Drafts, firms with Net Documents native workflows) get less from this than the headlines suggest.
That said, Microsoft shipping into Word added one option to one slice of one workflow. The diagnosis underneath it - push what you have to its ceiling first, name the gap specifically when you find it, then pick the cheapest move that closes it without breaking what works - is not to be neglected.
Legal AI in Action
📰 Interviewed for Law.com on the State of Agentic AI in Legal
Got included in Law.com piece Benjamin Joyner wrote last week on the failure modes that show up when AI agents interact with other agents and tools - the kind of things that have me telling boutiques to skip the autonomous moonshots for now and stay with defined AI workflows (same upsides without the downsides). Humbling company to be in alongside Tom Barnett (Maker5) and Will Gaus (Troutman Pepper Locke).
🎬 The Boutique AI Climb - Full Operational Walkthrough
Why intake → matter management → billing in that order, what the pre-brief actually looks like the moment a partner opens a matter, and the three places I’m still working out where the framework holds and where it doesn’t.
🎙 AI At The Operating Level
Jiyun Hyo is the founder of Givance, an MSO that runs the operational layer of a law firm (admin and paralegal work) so the lawyers can stay focused on legal judgment. We talked through why he stopped selling AI tools to AmLaw firms, why the work that doesn’t require legal judgment is where AI adoption actually pays off inside a firm, where agents like OpenClaw fit and where they break down once traceability matters, and what the 5 to 20 lawyer firm size lets a model like Givance do that the bigger firms can’t.
Coming up!
🎙 Next Tuesday at 2pm CET!
Next week’s guest on Rok’s Legal AI Conversations is Helen Fan, California attorney based in Silicon Valley, Chief AI Officer at a boutique law firm there, and the author of the Legal AI Value Stack, the framework that put a name on where defensibility actually lives in legal AI today.
We discuss her order of operations that a boutique firm can actually run, what she’s learning from running Open Claw LLP, her sandboxed AI-native experiment with two AI agents (senior and junior attorney) she built herself, and why she’s more excited about boutiques and mid-sized firms in the AI era than BigLaw, where the billable hour math cuts against AI adoption.
Each edition of Legal AI Brief brings practical lessons from firms using AI safely.