Harvey is among the companies adapting generative artificial intelligence to the requirements of a profession rather than offering a general-purpose assistant. Based in San Francisco, the American company primarily targets law firms and corporate legal departments. Its platform, presented on harvey.ai, aims to reduce the time spent processing documents and performing certain legal drafting tasks in a sector where confidentiality and reliability are critical.
Bringing together law and AI research
Harvey was founded in 2022 by Winston Weinberg, a former lawyer, and Gabriel Pereyra, an artificial intelligence researcher who previously worked at DeepMind and Meta, among others. This duo combines knowledge of legal practice with expertise in machine learning models. The company received early backing from the OpenAI Startup Fund, followed by investors such as Sequoia Capital.
Its visibility grew in 2023 with the announcement of its deployment at Allen & Overy, a firm now part of A&O Shearman. A partnership with PwC also demonstrated its potential beyond law firms alone, particularly in legal and tax activities. These agreements helped raise awareness of an approach focused on professional applications in law.
Tools for working on legal matters
The platform enables users to query documents, generate summaries, compare clauses and prepare drafts. It also supports legal research and the review of large volumes of documents, for example during acquisition due diligence. The aim is to integrate AI into existing workflows, with features tailored to each organization’s matters and information.
Harvey relies on large language models combined with specialized features and documentary sources. Its commercial proposition is based on software for organizations rather than a consumer service. Access management, data protection and the ability to verify the material used in a response are therefore essential criteria for its buyers.
These tools do not, however, eliminate the limitations of generative AI. A plausible response may contain an error, an inaccurate reference or an interpretation unsuited to the context. Review by a legal professional remains necessary, particularly when the output commits a client to a course of action or informs a decision.
What next?
For Harvey, the challenge is to turn initial deployments into sustained, measurable use integrated into clients’ systems. Its growth will depend in particular on its ability to cover different jurisdictions, document the quality of its results and meet local confidentiality requirements. In competition with established legal publishers and general-purpose assistants, differentiation will hinge on the relevance of sources, the depth of profession-specific features and user trust. Changes in the billing of legal services could also influence how productivity gains are valued.