The chess pieces moved fast in early 2023. Allen & Overy had just announced its exclusive Harvey deal—3,500 lawyers, splashy press coverage, the works. OpenAI was hours away from unveiling GPT-4 to the world. And then, on March 1, Casetext flipped the table.
CoCounsel launched that morning. Not a pilot. Not a limited beta. A commercial product, seven distinct legal skills, enterprise pricing, the whole infrastructure. Within hours—hours—Fisher Phillips had rolled it out to more than 500 attorneys. DLA Piper, which had been quietly testing since the previous September, went public with results. Ford Motor Company's legal team signed contracts.
It wasn't subtle. This was a land grab.
Three and a half months later, Thomson Reuters cut a check for $650 million.
The story of how Casetext executed that sprint reveals something about the current AI moment that most breathless coverage misses: being first matters less than being deployable. And for all the talk of GPT-4's capabilities, what actually sold wasn't the model—it was the scaffolding Casetext built around it.
The Pitch Was Deceptively Simple
Tasks that bill for hours get done in minutes. That was the core promise, and Casetext knew its audience wouldn't buy vague productivity theater. So CoCounsel shipped with specificity: upload a set of contracts, ask targeted questions, get back a grid matching your queries to relevant documents with drill-down justifications. Need a research memo? Feed in your facts, receive structured analysis with proper citations. Deposition prep? Here's your outline and a list of potential questions grounded in the actual case materials.
The interface told you everything about who designed this. No blank chatbox staring back, inviting the kind of open-ended hallucinations that make general counsels nervous. Instead: structured workflows. Pick a skill, provide inputs, receive formatted outputs. There was even a "Safety Mode" warning when users veered toward unstructured chat—a deliberate nudge back toward reliability over flexibility.
Casetext integrated CoCounsel with its existing Parallel Search technology and proprietary legal databases. Not just a language model wrapper, in other words. And the security pitch was there from day one: client data would never train the models, firms retained complete control of their information. Boring details, maybe. But the kind of boring details that actually get past enterprise procurement teams.
The company had spent 4,000 hours on training and fine-tuning across more than 30,000 legal questions. Four hundred attorneys in beta had generated over 50,000 uses before general availability. Tests ran into the thousands. Perhaps more than the founders initially expected, but this was the necessary tax for selling into a profession where a single mistake can end a career.
The GPT-4 Secret (That Wasn't Really Secret for Long)
On March 14—two weeks post-launch—Casetext confirmed what industry insiders had suspected: CoCounsel ran on OpenAI's GPT-4, the same model that had just been publicly announced. The company had secured early access, collaborated with legal scholars Dan Katz and Michael Bommarito on research showing GPT-4 could pass the Uniform Bar Exam. Around the top 10% on both multiple-choice and written components, the study found.
"GPT-4 leaps past the power of earlier language models," Pablo Arredondo, Casetext's chief innovation officer, said at the time. The bar exam performance became marketing collateral, though the company was careful—smarter, really—to frame CoCounsel's reliability around its structured approach rather than the raw model alone.
Smart move. Because by summer 2023, everyone would have GPT-4 access. The moat wasn't the model.
The Customer List Read Like an Am Law Directory
Fisher Phillips. Eversheds Sutherland. Orrick. DLA Piper. Frank Ryan, Americas Chair at DLA Piper, said the tool was "changing how the law is practiced by automating critical, time-intensive tasks." Darth Vaughn from Ford Motor Company's legal team called it technology with "the potential to revolutionize the way lawyers practice."
Marketing copy? Sure. But backed by actual deployments.
By launch, Casetext already had relationships with more than 10,000 law firms, including over 40 Am Law 200 shops—though that figure reflected the company's broader platform, not just CoCounsel adoption. Still. Distribution advantages matter when you're trying to move fast, and Casetext wasn't starting from zero. It was upselling an existing customer base that already trusted its research tools.
Third-party pricing data suggested CoCounsel All-Access ran around $500 per month per user, with on-demand options at $50-75 per task. For firms billing $300 to $1,000-plus per hour, the ROI math was straightforward. If the tool delivered.
Then Thomson Reuters Showed Up With a Check
June 26, 2023. Thomson Reuters announced a definitive agreement to acquire Casetext for $650 million in cash. The deal closed August 17.
Four months from launch to exit. In venture-backed tech, that's practically a sprint.
The acquisition wasn't purely about buying a hot AI product—though that certainly helped. It was defensive chess. LexisNexis and Thomson Reuters have long dominated legal information services, but generative AI threatened to unbundle those carefully constructed database monopolies. CoCounsel gave Thomson Reuters an offensive play in the new paradigm, a way to own the transition rather than watch it happen.
By November 2023, Thomson Reuters was weaving CoCounsel into its product suite: Westlaw Precision, Practical Law, Document Intelligence, HighQ. The tool became "CoCounsel Core." International rollout began—Canada and Australia in early 2024, the UK shortly after. Eight skills were listed in the expanded version, including a timeline feature that hadn't been in the original launch package.
A January 2024 press release claimed "six of the ten top US firms" were using CoCounsel. A post-acquisition metric, yes, and one that validated the land-grab strategy but also reflected Thomson Reuters' ability to drive adoption through its existing relationships and sheer market presence.
About That "First AI Legal Assistant" Claim
Casetext called CoCounsel "the first AI legal assistant," a phrase that appeared in press materials and executive quotes. It deserves scrutiny.
Allen & Overy had announced its Harvey partnership on February 15, 2023—two weeks before CoCounsel launched—with over 3,500 lawyers testing the tool since November 2022. PwC signed a global Harvey deal on March 15, two weeks after CoCounsel's debut.
So the "first" framing was marketing, the kind of carefully worded claim that looks good in headlines and doesn't quite hold up under examination. Several AI legal tools were moving simultaneously in early 2023. But Casetext had one advantage: a structured, task-specific product ready for broad commercial release while competitors were still running exclusive pilots with limited access.
Timing, in other words. And packaging.
LexisNexis eventually responded with Lexis+ AI on October 25, 2023, emphasizing hallucination-free citations and grounded analysis. The incumbents were awake, and they had distribution advantages Casetext couldn't match alone—hence, perhaps, the wisdom of that Thomson Reuters acquisition.
The market has since fragmented. A Law360 Pulse report in 2024 noted some early CoCounsel adopters had churned as competition intensified and firms tested alternatives. This is normal in emerging categories. Early adoption doesn't guarantee retention when buyers have options, and by 2024, buyers had plenty of options.
What Actually Mattered
CoCounsel's success wasn't about being first—not really, since they weren't. It wasn't purely about GPT-4 access, either, since that advantage evaporated within weeks.
What Casetext understood, and executed on with unusual clarity, was how professional service buyers actually make decisions. Lawyers needed structured workflows, not chat interfaces. Security guarantees, not vague promises about privacy. Concrete ROI stories, not hand-waving about productivity gains. And they needed a vendor they could explain to risk committees without triggering alarm bells.
Casetext delivered that package in March 2023, leveraged it into a $650 million exit four months later, and demonstrated—for anyone paying attention—that vertical AI applications can command enterprise premiums if they solve real workflow problems with enough specificity to get past procurement.
The broader lesson? Speed to market matters. But only if you ship something buyers can actually deploy. And sometimes the moat isn't the technology at all. It's understanding that the gap between "impressive demo" and "enterprise-ready product" is where most AI startups go to die.
Casetext didn't die there. They sprinted across it, grabbed the land they could, and sold before the next wave arrived.
That's one way to win.
