Law firms have a reputation for technological conservatism that their 2024 and 2025 AI adoption pace does not support. The same institutions that spent a decade resisting cloud document management and a further decade resisting alternative fee arrangements have moved faster on AI adoption than almost any comparable professional services sector. The explanation is not that law firms have changed their institutional culture. It is that the productivity differential between AI-assisted and unassisted legal work has become large enough to threaten client relationships, competitive positioning, and partner economics simultaneously, and that combination of pressures moves even conservative institutions quickly.
The economics that forced the pace
Legal work has a cost structure that makes AI’s productivity implications unusually transparent. Law firms bill by the hour. When AI reduces the time required to perform a task that previously billed at associate rates for multiple hours, the firm faces a direct choice: reduce the bill to the client, maintain the bill and absorb the productivity gain into margin, or find alternative work for the hours freed up. All three choices have been made by different firms, and the competitive pressure from firms that have passed productivity gains to clients is reshaping billing conversations across the market.
Clients have noticed. Large in-house legal departments, operating under budget pressure and with enough contract volume to run their own productivity benchmarks, have been explicitly asking their law firm panels about AI capability and asking whether AI-producible work is being billed at full hourly rates. The firms that can answer these questions credibly, by demonstrating AI capability and transparent billing practices, are winning mandates at the expense of those that cannot.
The result is an adoption curve that bypassed the usual slow evaluation phase. Firms including Allen and Overy through its Harvey AI partnership, Linklaters, Mishcon de Reya, and a wave of mid-market and regional firms moved from pilot to production deployment of AI legal tools in the 2023 to 2025 period at a pace that would have seemed implausible three years earlier.
Harvey AI and the full-stack legal platform
Harvey AI deserves specific attention because it represents a distinct approach to legal AI that has defined much of the sector’s recent development. Rather than building narrow-purpose tools for specific legal tasks, Harvey built toward a full-stack AI platform for legal work, capable of handling drafting, research, due diligence, contract review, and client communication tasks within a single system.
The architecture relies on a combination of general-purpose foundation models, fine-tuned on legal text and workflows, with a retrieval layer that surfaces relevant precedents, statutes, and firm-specific knowledge in response to each task. The result is a system that behaves more like a trained associate than a search tool: it produces outputs in the format and at the quality level that a legal workflow requires, rather than producing raw information that a lawyer must then structure into a usable work product.
Allen and Overy’s deployment of Harvey, among the first large-scale law firm deployments of a full-stack legal AI system, produced outcome data that accelerated adoption across the market: associate hours on AI-suitable tasks reduced substantially, and client satisfaction scores on the quality and responsiveness of work product improved rather than declining. The specific finding that quality improved rather than declined was the data point that moved the conversation at firms still in evaluation mode.
Due diligence: the highest-volume application and the clearest ROI
Legal due diligence, the systematic review of documents in connection with transactions, regulatory proceedings, and litigation, has historically been one of the most time-intensive applications of junior legal talent. A large M and A transaction might require review of hundreds of thousands of documents by teams of associates working extended hours over compressed timescales. The cost is substantial, the work is intellectually undemanding for the lawyers performing it, and the timeline pressure creates quality risks that manual review at speed generates.
AI document review has produced documented productivity improvements in this context that are large enough to have changed how law firms staff and price due diligence projects. Relativity and Everlaw on the litigation and regulatory review side, and Kira Systems, now part of Litera, on the transactional due diligence side, have each generated outcome data showing AI-assisted review reducing document review time by fifty to eighty percent on suitable document sets.
The qualifier “suitable document sets” matters. AI document review performs most reliably on large volumes of relatively standard documents where the review criteria are clearly specifiable. It performs less reliably on highly complex, multi-jurisdictional document sets where the relevance criteria are ambiguous or where documents require contextual interpretation that goes beyond pattern matching. The due diligence workflows that have achieved the highest AI productivity gains are those that have designed the human and AI task allocation carefully, reserving human judgment for the interpretive layer and using AI for the volume processing layer.
Legal research: transformation of a foundational workflow
Legal research, the identification and analysis of relevant precedents, statutes, and regulatory guidance, is the foundational workflow that underpins virtually all substantive legal work. AI has entered this workflow through two distinct vectors.
The first is enhanced search, where tools including Lexis Plus AI, Westlaw Precision, and Casetext’s CoCounsel use large language model capabilities to answer natural language research questions rather than requiring Boolean search syntax. The productivity gain from this interface change is measurable but incremental: lawyers who were already proficient in legal database search find AI-assisted research somewhat faster; lawyers less proficient in database search find it substantially faster.
The second is synthesis and analysis, where AI tools produce structured summaries of relevant authorities on a question, identify the strongest precedents for a position, and flag counterarguments that the research should address. This capability is more transformative than enhanced search because it changes what is possible in the research workflow, not just how fast existing tasks are completed. A research memo that previously required a junior associate two days to produce can be drafted in draft form in two hours with AI assistance, with the lawyer’s time then concentrated on evaluation and judgment rather than identification and collection.
The validity risks in AI legal research deserve honest acknowledgment. AI research tools can and do produce confident-sounding references to cases and statutes that do not exist, or that exist but do not stand for the proposition cited. The well-publicized incidents of lawyers submitting AI-generated briefs with fabricated citations established empirically that this risk is real and consequential. The workflow governance required to use AI legal research safely is specific: treat AI research outputs as a starting point requiring verification, not as a final product.
The regulatory position law firms operate in
Law firms face a specific tension in AI adoption that other professional services firms do not share in the same form. The duty of competence, the professional obligation to remain current with relevant technology, has been interpreted by several bar associations as requiring lawyers to understand the AI tools they use and to use them competently. The duty of confidentiality imposes constraints on routing client information through third-party AI systems that do not provide adequate data handling guarantees. The prohibition on unauthorized practice of law creates questions about AI-generated legal analysis that goes beyond workflow assistance into substantive legal advice.
The firms that have navigated this regulatory environment most effectively have not waited for definitive bar guidance that may not arrive for years. They have built their AI governance frameworks on conservative interpretations of existing professional obligations, with legal counsel review of AI system contracts and data handling practices, and with deployment governance that maintains clear human accountability for all substantive work product. This approach produces slower initial deployment than a permissive interpretation would allow but more durable compliance architecture than ad hoc responses to professional conduct inquiries.
Law firm AI adoption has moved faster than the sector’s reputation for conservatism would predict, and the driver is economic rather than cultural: the productivity differential has become large enough to affect client relationships and competitive positioning in ways that institutional inertia cannot resist. The firms that are building AI capability deliberately, with governance frameworks appropriate to their professional obligations, are generating sustainable productivity advantages. Those adopting AI reactively, under competitive pressure and without adequate governance, are creating liability exposure that the conservatism they departed from was designed to prevent.
For the contract management dimension of legal AI, see AI in contract management: automating legal work. For the broader legal technology stack transforming workflows, read legal tech AI: the tools redefining legal workflows.
The question managing partners and general counsel should ask about their AI programs: Which professional obligations does your AI governance framework specifically address, and has that framework been reviewed by someone who understands both the technology and the applicable professional conduct rules?
