In the span of a few years, legal AI has gone from an experimental curiosity to one of AI's richest vertical markets. Harvey is its most evident symbol. The company founded by Winston Weinberg and Gabe Pereyra has announced a new $550 million funding round at a $15.5 billion valuation, co-led by Diffusion and Lightspeed Venture Partners. It is the third major jump in valuation in less than a year, bringing the total capital raised to over $1.5 billion.
The figure is impressive, but the most interesting aspect is not financial. Harvey claims to be used by 80 percent of Am Law 100 firms and five Fortune 10 companies. Legal AI is therefore entering the organizations that have the most to lose from an error, a data leak, or an incorrect citation.
From chatbots to proprietary knowledge
The first phase of professional AI consisted of querying a large generalist model. Harvey is attempting to shift the center of gravity toward systems trained and adapted for the legal domain and, above all, toward a model where clients themselves can own a greater share of their application intelligence.
The company recently unveiled Tenet, its first post-trained model based on open weights, and Harvey LAB, a benchmark dedicated to legal agents. The move is significant because it challenges the notion that value must necessarily reside in the largest foundation model. In law, the competitive edge may lie in data quality, workflows, verification, and firm-specific knowledge.
Why lawyers have become an ideal market
Legal work involves a massive amount of high-value text: contracts, briefs, precedents, due diligence, compliance, and corporate filings. A substantial portion of professional time is spent searching, comparing, synthesizing, and producing documents. These are activities where language models can drastically increase speed, but where an error has tangible consequences.
This makes the sector both ideal and demanding. The product cannot merely generate a plausible answer: it must integrate into document management systems, respect permissions, maintain audit trails, and enable professionals to verify the work.
A valuation betting on distribution
Harvey is now worth far more than numerous software companies with established revenues. Investors are implicitly betting that whoever wins over top law firms first can become the operating layer through which an increasing share of legal work will flow.
It is the same dynamic that made enterprise ERP systems so massive: once integrated into mission-critical processes, replacing them becomes costly. In AI, this effect can be even stronger if the platform accumulates client-specific configurations, knowledge, and procedures.
Open models change reliance on AI labs
The decision to also work with open-weight models further points to a possible future for vertical AI. Specialized companies might not need to build multi-billion-dollar foundation models from scratch, but can instead take an existing base and transform it through domain-specific data and post-training.
If this approach succeeds, industries such as law, finance, medicine, and engineering could develop intelligence stacks that are increasingly autonomous from major generalist labs.
The risk of speed
However, rapid growth brings an equally swift responsibility. A system used for due diligence, litigation, or compliance must be evaluated not just for average quality, but for edge cases. Human oversight remains essential, as does the distinction between automating a task and transferring professional liability.
Harvey is not proving that lawyers will become obsolete. It is demonstrating something perhaps more important: an extremely conservative profession has begun to view AI as a standard component of its infrastructure. When this happens, technological transformation stops being a forecast and becomes an everyday organizational challenge.
