Full profile
The useful Harvey AI law question in 2026 is no longer whether large firms are experimenting with generative AI. They are. The sharper question is where the tool has entered daily legal work, and whether firmwide access has translated into repeated lawyer use.
The current evidence is deliberately mixed. In the SKILLS 2026 survey, fielded in Q4 2025 across 130 of the largest global law firms, Harvey led seven of eleven tracked legal AI use-case categories: legal drafting, contract review, due diligence, contract negotiation, playbook generation, discovery automation, and timelines.[1] In the same market picture, firms were running an average of 18 live AI solutions, while only about 20% of lawyers at the largest firms were using AI legal assistants regularly.[1]
That is the gap worth taking seriously. Category leadership says something real about enterprise selection and practice-group experimentation. The 20% figure says something equally real about the slower work of changing how lawyers draft, review, negotiate, and prepare matters under time pressure.
A source caveat belongs near the front. The SKILLS findings discussed here were published on Harvey’s site, and Harvey has an obvious interest in emphasizing favorable results. SKILLS founder Oz Benamram discussed the survey in a Harvey-hosted Q&A, which helps clarify the survey frame but does not turn the publication into independent vendor-neutral research.[2] Artificial Lawyer’s separate coverage of which legal AI tools firms are actually using gives partial external validation that the survey findings are being discussed beyond Harvey’s own marketing channel.[3] The safest reading is not “independent proof of universal adoption.” It is “a useful, named-workflow data point from the largest-firm market, with vendor-hosted presentation.”
Where Harvey Shows Up In The Work
The seven categories matter because they are not vague claims about “productivity.” They name pieces of legal work that already have owners, templates, review habits, risk thresholds, and client expectations. If a tool is going to survive past the launch memo, it has to fit into those pieces of work rather than ask lawyers to invent a new workday around a chat window.
| SKILLS category led by Harvey | What the category usually means inside a large firm |
|---|---|
| Legal drafting | Producing, revising, and structuring legal documents from instructions, precedents, or matter materials |
| Contract review | Checking contract language against issues, policies, fallback positions, or commercial instructions |
| Due diligence | Reviewing document sets to extract obligations, risks, change-of-control issues, or other matter-specific findings |
| Contract negotiation | Comparing proposed language with preferred positions and preparing responses or revisions |
| Playbook generation | Turning firm or client positions into reusable guidance for repeated review and negotiation |
| Discovery automation | Helping manage litigation document review, relevance analysis, and related discovery tasks |
| Timelines | Building chronologies from matter records, correspondence, pleadings, or evidence |
The distribution is telling. Five of the seven categories cluster around transactional execution: drafting, contract review, due diligence, negotiation, and playbook generation. That does not mean litigation use is marginal. It does suggest that Harvey’s strongest large-firm footprint, at least in this data set, sits where work can be decomposed into repeatable document-heavy steps.
Transactional Work Is The Center Of Gravity
In transactional matters, the attraction is not that AI “writes contracts.” That phrase is too broad to be useful. The more credible version is narrower: lawyers use Harvey in tasks where there is already a body of precedent, a known transaction structure, a set of client positions, and a review path that ends with a human lawyer taking responsibility.
Drafting is the obvious starting point, but not necessarily the most mature one. A senior associate may need an initial clause, a comparison against precedent language, a revised recital package, or a first pass at a document section that still has to be checked against deal facts. The workflow value is not that the lawyer stops drafting. It is that the first movement from instruction to working text becomes faster, more structured, and easier to interrogate.
Contract review is more operational. A team can ask whether a document follows a playbook, whether a clause departs from a preferred position, or whether a counterparty’s markup creates a risk that belongs in the negotiation tracker. The work still lands with lawyers, but the tool can reduce the amount of page-by-page searching before a judgment call is made.
Due diligence is where repeatability becomes especially important. The relevant question is not whether an AI system can summarize documents in the abstract. It is whether the review team can define the issue list, process large volumes of material consistently, surface exceptions, and produce outputs that a supervising lawyer can trust enough to include in a diligence report or risk discussion. Harvey’s leadership in due diligence in the SKILLS categories therefore points to a practical area of deployment rather than a general endorsement of AI summarization.[1]
Negotiation and playbook generation sit together. A playbook is not merely a document stored in a knowledge folder. Used properly, it is the firm’s or client’s position architecture: what is acceptable, what requires escalation, what fallback language can be offered, and which positions are commercial rather than legal. If Harvey is being used to generate or apply those playbooks, the work moves closer to institutional knowledge management than one-off AI queries.
This is why the transactional categories deserve more attention than a generic “legal drafting” headline. Large firms do not win adoption simply by giving every lawyer an empty chat box. They have a better chance when a banking group, private equity team, funds practice, or commercial contracts unit can connect the tool to the documents and decisions that recur across matters.
Litigation Use Is Real, But The Pattern Is Different
The SKILLS categories also place Harvey in discovery automation and timelines.[1] Those are litigation-native workflows, and they have a different adoption profile from transactional drafting. Litigation teams are already accustomed to specialized tools for document review, e-discovery, deposition preparation, case chronology, and research. Harvey enters a more tool-saturated environment.
Discovery automation is not just “read these documents.” In practice, it may involve identifying potentially relevant material, clustering issues, preparing summaries, helping reviewers move through document sets, or assisting with privilege and responsiveness workflows under supervision. The margin for error is shaped by court obligations, production strategy, privilege risk, and opposing-party scrutiny. That makes governance as important as speed.
Timelines are a natural companion use case because litigators spend so much time turning messy records into sequence: who knew what, when a notice was sent, how a contractual obligation developed, when a representation was made, and which document supports each point. A chronology that cannot be traced back to source material is not very useful. A chronology that helps the team find and test the source trail can save meaningful time before witness preparation, pleading amendments, mediation statements, or trial preparation.
Harvey’s use-case material also points to litigation workflows such as case law research, brief drafting, and trial preparation as part of the broader pattern of platform activity. Those should be read carefully. The strongest cited SKILLS leadership categories here are discovery automation and timelines, not a blanket claim that Harvey dominates every litigation task. The narrower point is enough: the platform is appearing in litigation workflows where document volume, chronology, and drafting pressure make structured assistance valuable.
Workflow Agents Show The Deployment Mechanism
The most interesting Harvey metric may not be an adoption percentage. Harvey says users have built more than 25,000 custom Workflow Agents across more than 500 practice groups.[4] Because this is Harvey’s own disclosure, it should not be treated as an independent measure of effectiveness. Still, the metric points to the right enterprise question: are firms encoding their own legal and procedural knowledge, or are they simply giving lawyers access to a general assistant?
A custom workflow agent can be valuable because law firm knowledge is rarely just “the law.” It includes preferred drafting positions, client-specific risk tolerances, partner comments that have become house style, local filing habits, escalation rules, precedent selection, and the practical difference between a point that can be conceded and a point that must be discussed with the client.
This is also where deployment becomes labor-intensive. Someone has to decide which precedents are approved. Someone has to test outputs against real matter facts. Someone has to monitor whether lawyers are using the workflow as designed or bypassing it when the deadline tightens. The 25,000-plus agent figure is encouraging because it suggests actual configuration work. It is not, by itself, evidence that every agent is good, governed, current, or widely used.
Firmwide Availability Is Not The Same As Lawyer Adoption
The roughly 20% regular-use figure is the finding that should stay on the desk after the demo ends.[1] It is not a gotcha. In large firms, a technology can be licensed, announced, trained, integrated with security requirements, and still not become part of daily practice for most lawyers. That is especially true when the work product is client-facing and the professional risk remains with the lawyer, not the software.
There are at least three different maturity states that often get collapsed into “adoption.”
- Firmwide deployment: the platform is available, approved, and technically usable by a broad population.
- Practice-group experimentation: specific teams test the tool against drafting, review, diligence, research, or litigation workflows.
- Regular individual use: lawyers return to the tool repeatedly because it fits the matter rhythm better than the old combination of Word, Outlook, search, and precedent banks.
The SKILLS numbers imply that many large firms are somewhere between the second and third states. They have multiple live AI tools, and Harvey appears to be leading several meaningful categories. But regular individual use remains limited to a minority of lawyers.[1] That is the operational constraint innovation leaders have to manage: not procurement, but conversion of available capability into reliable matter behavior.
The gap also helps explain why task specificity matters. A lawyer under pressure is unlikely to open a general AI assistant just because it exists. The lawyer is more likely to use a workflow that answers an immediate matter need: compare this draft to the playbook, extract the change-of-control provisions, turn these documents into a chronology, prepare a negotiation position, or produce a first draft from approved precedent. Adoption grows from repeated usefulness at those points, not from enterprise enthusiasm.
In-House Metrics Need A Separate Reading
Harvey reports high monthly adoption figures at customer organizations, including 95% at CMS and 96% at Repsol, and states that in-house users save more than 25 hours per month per lawyer.[4] These are useful claims, but they are vendor-attributed customer metrics. They should not be read as independently audited proof that every comparable legal department will see the same pattern.
They also describe a different operating environment from a global law firm. Corporate legal departments may have more centralized legal operations control, more standardized intake, repeat contract portfolios, and clearer reporting lines for tool usage. A legal department reviewing a recurring class of commercial agreements can often make a workflow mandatory in a way that a multi-practice law firm cannot.
That does not make the in-house figures irrelevant. It means they should be used to ask better questions. Which matters produced the reported time savings? Were the hours saved concentrated in contract review, regulatory tracking, playbook use, or another workflow? Did lawyers use the tool directly, or did legal operations teams structure the process around them? Without those details, the numbers indicate potential scale, not guaranteed transferability.
What Buyers Should Take From The Gap
For large firms and in-house teams, the practical lesson is not to discount Harvey because regular individual-lawyer adoption is around 20%. Nor is it to treat category leadership as proof that workflow change has already happened. Both readings are too easy.
The better reading is that Harvey has credible evidence of production relevance in named legal workflows among the largest firms, especially in transactional categories where drafting, review, diligence, negotiation, and playbook work can be structured. At the same time, the deployment-adoption gap remains the central management problem. A firm can buy the platform. It cannot buy, in the same transaction, partner confidence, associate habit change, approved content libraries, matter-specific supervision, or client-aligned governance.
This is also not a universal legal AI buying guide. Harvey’s current evidence base is strongest as a large-firm and in-house story. Small and solo firms have different constraints: budget, volume, IT support, risk tolerance, and implementation capacity. The SKILLS survey population of the largest global firms should not be stretched into a conclusion about the whole legal market.[1]
There is one important unknown the available sources do not resolve: whether the regular individual-use rate is rising, flat, or unevenly distributed by practice group. No cited source provides independent longitudinal data on that point. For 2026 planning, that absence matters. A buyer should want to know not only whether a platform can be launched, but which groups keep using it after the novelty has passed.
Harvey’s current evidence supports category leadership in large-firm legal AI deployment. The roughly 20% regular lawyer adoption figure remains the operational constraint to watch. It is not proof that Harvey is failing, and it is not proof that legal AI has arrived everywhere. It is the normal but consequential distance between enterprise rollout and embedded workflow change.
The decision-maker’s question for 2026 is therefore narrower and more useful than “Does Harvey have customers?” or “Can Harvey perform legal tasks?” It is: which practice groups will actually build, govern, and repeatedly use these workflows once the platform is live?
References
- 2026 SKILLS Survey: Where Legal AI Is Working, Harvey
- 2026 SKILLS Legal AI Survey: Q&A With Oz Benamram, Harvey
- Which Legal AI Tools Are Law Firms Actually Using?, Artificial Lawyer, June 2, 2025
- Top Harvey Use Cases, Harvey
Comments
Join the discussion with an anonymous comment.