Harvey Builds the Intelligence Layer for Legal Work

Legal knowledge is expensive because the useful part rarely sits in one document. It lives across precedents, redlines, matter files, client history, partner preferences, and the judgment required to know which detail controls the outcome. Harvey is building an AI platform designed to make that institutional intelligence usable without separating it from the permissions, citations, and professional review that legal work demands.

Founded in San Francisco in 2022 by CEO Winston Weinberg and President Gabe Pereyra, Harvey serves law firms, in-house legal teams, and professional-services organizations. Weinberg brought experience as a securities and antitrust litigator at O'Melveny & Myers. Pereyra brought machine-learning experience from DeepMind and Meta. That pairing gave Harvey a useful starting argument: legal AI needs both technical ambition and respect for how professionals actually produce work.

Harvey matters now because the company is moving from an assistant toward an intelligence layer for high-stakes professional workflows. The company reported more than $400M in annual recurring revenue and more than 3,000 customers in September 2026, including 80% of the Am Law 100, 20% of the Fortune 500, and five of the Fortune 10. Those figures are company-reported, but they show why Harvey has become one of the clearest tests of whether vertical AI can earn enterprise trust at global scale.

About Harvey

Harvey began after Weinberg and Pereyra experimented with ChatGPT and recognized that legal work offered a demanding proving ground for generative AI. The founders cold-emailed OpenAI, which became an early investor, and built the company around domain-specific legal and professional workflows rather than a general consumer chatbot.

The product has expanded into a connected platform. Assistant supports research, analysis, drafting, and source-linked answers. Vault organizes large document sets and turns extracted information into review tables. Workflow Agents automate repeatable work, while Agent Builder lets organizations encode their own playbooks, standards, and methods. Spaces gives firms, clients, and internal teams a governed place to collaborate on matters and shared work product.

The important design choice is connection. A legal team does not need five clever demos that lose context at every handoff. It needs documents, knowledge, workflows, approvals, and outputs to stay attached to the same matter. Harvey is trying to make that continuity a product advantage.

Why Harvey Matters Right Now

Harvey's scale is beginning to change the technical problem. The company says customers now run more than 700,000 agent tasks each day, extract 50 million terms weekly, and analyze more than 5.8 million documents daily. Those workloads create pressure to improve quality, latency, cost, observability, and governance at the same time.

That pressure explains Harvey's investment in specialized models and evaluations. Harvey Tenet is a post-trained open-weight model built for long-horizon legal work, while Harvey LAB tests agents against realistic client matters and expert rubrics. The broader bet is that a vertical AI company can do more than route requests to whichever frontier model is newest. It can build the evaluation environments, product controls, and domain infrastructure that make a model useful inside a profession.

Harvey's latest $550M financing at a $15.5B valuation gives that bet more room and a much higher execution bar. Capital can fund compute, research, acquisitions, and global hiring. It cannot manufacture the professional confidence required before an agent touches a live matter.

The Trust Layer Is Part of the Product

Legal AI carries a different failure budget than a casual writing tool. A weak citation, permission mistake, or unsupported conclusion can affect litigation, contracting, compliance, or a client relationship. Harvey's product strategy therefore puts security and review mechanics beside model capability.

Harvey says customer inputs, outputs, and documents are not used to train underlying models. Its security program includes encryption, role-based access, audit logs, data-retention controls, regional processing, ethical-wall enforcement, and contractual restrictions on model providers. The platform also keeps professional judgment in the loop through citations, review steps, approval controls, and audit trails.

Those controls are not decorative enterprise furniture. They determine whether an agent can move from a pilot group into the daily work of a global firm. Harvey's real competition is partly technical, but it is also operational: can the system remain predictable as more people, matters, data sources, and autonomous steps enter the workflow?

Leadership, Culture, and Hiring Momentum

Weinberg and Pereyra still anchor Harvey's product and market story, while CTO Siva Gurumurthy leads the engineering organization through the demands of enterprise scale. Gurumurthy previously served as CTO at Motive and as a director of engineering at Twitter. Harvey has also expanded its executive bench across operations, product, security, strategy, marketing, finance, and revenue.

The company's careers page names three values: Simplicity, Decisiveness, and Job's Not Finished. COO Katie Burke wrote in August 2026 that Harvey was hiring across most teams, with engineers and account executives likely to remain recurring needs. That hiring signal fits the work ahead. Specialized agents require researchers and engineers, global enterprise distribution requires legal and commercial operators, and customer adoption requires teams that can turn a platform into repeatable practice.

Harvey describes its culture as intense, which deserves to be read plainly rather than translated into recruiting confetti. The company is scaling in a market where models, customer expectations, and competitive pressure all move quickly. The opportunity is large, and so is the organizational load.

What Harvey Signals for Enterprise AI

Harvey's trajectory suggests that the most durable vertical AI companies may own more than a polished interface. They may combine distribution, workflow context, permission systems, specialized evaluations, institutional knowledge, and enough model control to improve quality and economics for a specific profession.

The moat is not a single answer. It is the system that can turn a firm's prior work into governed, testable, reviewable output without turning confidential knowledge into somebody else's training data. If Harvey can keep that promise while expanding from individual assistance into coordinated agents, legal teams may begin to treat AI less like rented intelligence and more like infrastructure they can shape.

That is why Harvey is worth watching beyond legal technology. The company is testing whether application-layer AI can become the place where professional judgment is organized, encoded, and deployed. The next phase will be measured not by how confidently the software writes, but by how reliably institutions can trust it with work that carries consequences.