What Harvey is doing to win the legal AI race it inadvertently started

Business Insider

13-05-2025

Legal tech was long a space that investors ignored. Then came Harvey.
In just three years, the company, which builds software for analyzing and drafting documents using legally tuned large language models, has drawn blue-chip law firms, Silicon Valley investors, and a stampede of rivals hoping to catch its momentum. Harvey has raised over half a billion dollars in capital, sending its valuation soaring to $3 billion.
On a recent Monday afternoon at Harvey's Manhattan office, I met cofounder and chief executive Winston Weinberg in a polished conference room named for Atticus Finch, a beloved character from the novel "To Kill a Mockingbird." I mentioned that since I started covering legal tech a month ago, my inbox has been flooded with pitches from legal-tech startups eager to explain how they're not Harvey.
Weinberg let out a soft chuckle. "I'll take that as a compliment, I guess."
In recent years, more competition has been encroaching on Harvey's territory, and fast. Hebbia, a knowledge-search platform, has made a more focused push into the legal sector, and Legora, which offers an AI-powered workspace where lawyers can draft, edit, and collaborate, is gaining traction with Harvey's core clientele of Big Law firms.
While legal tech was once the domain of ex-lawyers building tools for their peers, now it's attracting classically trained software engineers, eager to compete in a space without a staid market leader.
Harvey may have cracked the market open, but now it faces the classic innovator's dilemma: the very proof of concept it offered to the legal world is fueling a growing list of competitors.
The legal-tech land grab is on. The question is whether Harvey can maintain its first-mover advantage, or if it's simply cleared the path for the next breakout.
Want in? Get in line.
Part of what fueled Harvey's ascent is its go-to-market strategy. Early on, the company bet that winning over the country's most elite law firms would create a ripple effect across the industry.
It gated access to the product using a waitlist, allowing it to tightly tailor the tool to pilot users.
So far, its Big Law wager appears to be paying off. Weinberg said lawyers at eight of the 10 highest-grossing firms in the nation are now using Harvey.
The company tells Business Insider it crossed $70 million in annual recurring revenue last quarter, putting it on pace to smash its goal of $100 million ARR for the year.
"Once a subset of the market standardizes on a solution, it's kind of the solution," said Ilya Fushman, a venture capitalist who led Kleiner Perkins' investment in Harvey in 2023.
Leaning forward in a high-back, caramel-colored leather chair, Weinberg seemed unfazed by the growing competition.
Harvey's edge, he argued, lies in two places its rivals can't easily replicate: top-tier talent and a product strategy built on deep collaboration with its customers.
The 260-person startup has lured dozens of trained lawyers off the gilded path to Big Law partnership, offering stock options and a shot at shaping the future of legal practice.
To keep its edge, Harvey just made a key hire. Stripe veteran John Haddock has joined as chief business officer after a decade scaling one of Silicon Valley's most closely watched startups.
Haddock told BI he spoke to dozens of Harvey customers before accepting an offer. His decision boiled down to their love of the product. He called it Weinberg's "No. 1 maniacal focus."
"The best thing we can do is stay focused on: are we building stuff that lawyers really need and need every day?" he said. "Everything else takes care of itself."
Harvey goes multi-model
Harvey has been fighting the competition with one hand tied behind its back. Since its founding, the company has partnered closely with OpenAI to build custom models for lawyers. Its entire product ran on OpenAI's models.
It's a limitation that Harvey's rivals have been quick to point out. They argue their products are superior because they can cherry-pick from the best of Anthropic, Meta, or Google, depending on the task.
Now, Harvey wants to neutralize that criticism. The company tells BI it's going multi-model, starting with Anthropic's Claude and Google's Gemini.
Weinberg said Harvey didn't avoid other models out of loyalty to OpenAI, but necessity. Until recently, most major law firms would only approve AI tools that ran through Microsoft Azure, which meant models like Claude and Gemini couldn't clear security reviews.
Those constraints are falling away as vendors like Anthropic build the features enterprises demand and gain clout.
The move may also suggest Harvey is adapting to a clientele that's perhaps more opinionated about which models power their tools, especially as rivals pitch flexibility as a selling point.
Harvey's secret sauce
In a market where model performance offers marginal gains, Harvey is betting that its true edge lies in how deeply the product molds to the client.
"What I think is closer to a traditional moat," Weinberg said, "is we are very focused on customization, massively."
The company partners directly with firms to build bespoke legal workflow software. With A&O Shearman, for example, Harvey helped develop a merger control tool that taps the firm's global antitrust bench. For another client focused on private equity, the company built out deal-specific workflows.
Lawyers across those firms are using the tools, and the firms are selling that customized software to clients and other law firms, sharing a cut of the revenue with Harvey.
If customization is the moat, the obvious question becomes: how does it scale? One investor in a Harvey competitor asked at what point the company becomes overrun with service requests and support tickets.
The company's bet is that it can turn workflows from custom projects into reusable building blocks — a sort of Lego kit it can adapt for each new client.
It's a bold strategy, but in a crowded field, Weinberg believes that winning won't come from better answers. It'll come from building a system that grows with the people asking the questions.
"At the end of the day," he said, "what you want to do is build a solution that tracks the evolution of law over time."