If you were building the ideal AI tool for transactional lawyers from scratch, you’d want it to exist where lawyers conduct their day-to-day work, you’d want it to understand their language, and to help (not hover) during every contract draft and negotiation. That’s the premise behind Spellbook, the legal AI copilot that operates natively inside Microsoft Word. It’s built for transactional lawyering, not just document summarizing, and it’s fast becoming the platform of choice for teams that want to draft smarter, redline faster, and negotiate like they’ve got a dozen extra hands.
Spellbook’s approach is grounded in what legal work actually looks like, minute by minute: toggling between redlines, referencing precedents, responding to client emails, checking clauses against market norms, and mentally tracking a dozen risk vectors. Their answer? A one-stop AI that not only sees what you see but anticipates your next move, without breaking your focus or your formatting.
Why Word, Why Now
Launched in 2022, Spellbook was the first generative AI tool purpose-built for lawyers. Today, it’s used by over 4,000 legal teams in 80+ countries and has reviewed more than 10 million contracts. Its co-founders Scott Stevenson, Daniel Di Maria, and Matt Mayers built it on a simple premise: lawyers aren’t looking for a new ecosystem; they simply want a new tool that sits seamlessly inside the environment they already use every day.
That core philosophy has shaped Spellbook into a deeply embedded AI layer within Microsoft Word. It’s not trying to replace your drafting software, it’s trying to turn it into a superpower. Whether you’re starting from scratch or reviewing a redline-heavy 60-pager, Spellbook can draft, analyze, and benchmark in real time. And because it recognizes clause types, party names, deal context, and tone, the output doesn’t feel generic, it feels familiar.
This week, The Legal Wire interviewed Co-founder Scott Stevenson to unpack how a brand new AI copilot is reshaping contract work from inside lawyers’ safe spade: Microsoft Word.
TLW: What early lessons shaped your decision to go all-in on Microsoft Word instead of building a standalone platform?
- “Using Github Copilot as an engineer–felt like getting an “electric bicycle”. I was still pedaling and steering, but it became 10x easier to go up hills. Much easier than going to a separate tool.” and
- “We watched lawyers try to adopt other products and fail. They were sick of jumping between apps.”
Legal Breadth, Precision Depth
Spellbook’s ambitions are wide-ranging: drafting, reviewing, benchmarking, playbooks, strategy. But with scale comes a challenge, and how do you expand your toolset without diluting legal accuracy?
That question becomes especially relevant with the introduction of Spellbook Associate, an AI agent designed to tackle multi-document workflows. Unlike a chatbot, Associate operates like a junior lawyer: it connects the dots across documents, updates terms at scale, and handles due diligence, data rooms, and disclosure schedules with surprising fluency.
TLW: What kinds of firms are finding the most value in Associate: large teams with volume, or lean teams needing leverage?
Scott: “Associate is being adopted by companies and law firms across the spectrum. It’s adopted by Fortune 1000 companies who need a better way to draft packages of documents (eliminating busy work), by large law firms doing M&A (speeding up lengthy processes and reducing risk in diligence) and used by small law firms in routine matters (simply helping small firms get things done with less support staff).”
Training the AI to Think Like You
Spellbook’s edge isn’t just its raw power (though it does run on OpenAI’s latest models. It’s how that power is tamed. Tools like Clause Libraries, Playbooks, and especially Preference Learning mean the platform continues to learn after its response.
Preference Learning is particularly compelling: the more you work in Spellbook, the more it adapts to your style. It picks up on your redline strategies, risk thresholds, and the clauses you favor, not just across documents, but across time.
TLW: What have you learned from users about how Spellbook evolves with daily use and what surprised you most about how lawyers personalize it?
Scott: “The main thing that surprised us is simply learning how individual every lawyer’s preferences are. In the world of AI benchmarking, there is often an assumption that there are objectively correct answers. There rarely are in legal.”
From Gut Check to Benchmarking Engine
Traditionally, figuring out if a clause was “market” meant calling a colleague or digging through old docs. Spellbook Benchmarks automates this with real-time comparison against more than 2,300 contract types. Of course, it flags differences, but it also explains why they matter and offers fixes on the spot.
Combine that with its integration into Thomson Reuters Practical Law, and you’ve got immediate access to trusted precedents without leaving Word. This isn’t generating AI hallucinating legalese, it’s citing from sources your clients (and regulators) respect.
TLW: How do you think about balancing AI assistance with regulatory and ethical expectations, especially when integrating third-party legal content?
Scott: “We think grounding, or retrieval augmented generation (RAG), are really superior techniques for getting models performing in specific verticals, because they allow you to actually cite and inspect realtime data sources. We think it’s super important to design products like Spellbook to have their output reviewed by the user, and to make it easy for them to inspect the trail of reasoning.”
Security, Scale, and Serious Funding
As Spellbook moves upmarket, it’s checking all the enterprise boxes. SOC 2 Type II, GDPR, CCPA, and zero data retention policies are all in place. And adoption is following suit. In October 2025, Spellbook closed a $50 million Series B led by Keith Rabois at Khosla Ventures, bringing its valuation to $350 million.
Half of Spellbook’s revenue now comes from large law firms and enterprise clients. That shift is influencing the roadmap: deeper CLM integrations, richer document storage, more robust administrative tools, and, crucially, support for more complex transactional work.
TLW: As enterprise adoption grows, how do you decide which features to prioritize for larger clients without losing the simplicity that attracted smaller firms?
Scott: “This is a tough balance, but we are obsessed with simplicity and low time to value. A lot of what we add is hidden behind the scenes, so that new complexity isn’t stacked onto the UI:
- Better data sources and integrations, so the AI can find more relevant info
- Better models and methods so that the AI is more performant
- Better, smoother user experiences
Making the product better doesn’t necessarily mean adding more complicated surface area to the product. We really try not to do that as much as we can.”
So What’s Next?
There’s plenty of buzz in legal AI, and throughout the buzz, Spellbook has remained sharply focused. On contracts, on Word, on lawyers. It’s avoided the trap of chasing every new use case and instead deepened its offering for transactional teams. It’s also managed to avoid the uncanny valley of generative legal tech: the results don’t feel off. They feel practical. Useful. Time-saving. (Sometimes even sanity-saving.)
TLW: You’ve described Spellbook as a copilot, but it’s increasingly doing the work of a junior associate. Where do you draw the line between AI assistance and automation in a profession built on human judgment?
Scott: “AI will certainly eat up more of the busy work as it progresses. But no lawyer I know enjoyed the work of a junior associate.
I think what’s most important is that these tools force lawyers to review the outputs–everything we output should just be a “suggestion” that can be accepted or rejected. This is true for both our copilot and Associate product.”
