Is Harvey better than a lawyer today?

Is Harvey AI actually better than a human lawyer today?

Harvey AI is already faster than human lawyers at several routine legal tasks, but it still cannot match an experienced attorney across the full range of legal work.

The difference becomes clear when we separate the work lawyers do. Reading hundreds of contracts, finding clauses, comparing documents, and producing first drafts are activities where AI has made extraordinary progress. Negotiating a difficult settlement, evaluating conflicting evidence, or deciding how much legal risk a company should accept requires a different level of judgment.

Harvey, founded in 2022 by Winston Weinberg and Gabriel Pereyra, has built a platform specifically for legal research, drafting, document review, and increasingly complex workflows.

The company is now worth $15.5 billion following a $550 million funding round in September 2026. According to Harvey, 80% of America's 100 highest-grossing law firms already use its platform.

But there is a surprising gap between that commercial success and what current AI systems can do independently. In Harvey's Legal Agent Benchmark, even the strongest model initially completed only 14.2% of complex legal assignments perfectly under a strict evaluation standard.

That leaves us with an interesting situation: law firms are spending heavily on technology that can transform their operations, even though they still need lawyers to check its work and take responsibility for the results.

Why are the world's biggest law firms adopting Harvey so quickly?

Harvey is spreading through major law firms because it helps lawyers complete more work without hiring proportionally more people.

The scale is remarkable for a company founded only four years ago. Harvey's September 2026 announcement confirmed that 80 of the Am Law 100 firms use its platform, alongside five of the Fortune 10 companies.

Its customer base is also becoming broader than traditional law firms. Harvey serves corporate legal departments, financial institutions, and professional services organizations that regularly deal with large amounts of legal documentation. Corporate legal teams and banks are buyers we profile, with the problems they pay to solve, in our legal tech market report.

Sacra reports that Harvey serves more than 1,500 customers in over 60 countries. According to its research, the platform has reached more than 142,000 lawyers.

One particularly interesting figure is how customers expand their usage. Sacra reports that Harvey's median customer doubles its number of licensed seats within 12 months. The same research found that weekly active users increased fourfold over a year, while monthly queries rose 5.5 times.

Those figures suggest that existing customers are getting employees to use the product more frequently, rather than simply maintaining subscriptions they barely touch.

There is still an important limitation: Harvey has not publicly provided enough consistent, independently verified data to calculate how much each customer saves or how often its work needs correcting.

But the repeated expansion of paid usage is difficult to dismiss. Law firms appear to be finding real value in the software.

Can Harvey finish a legal assignment without a lawyer checking it?

Harvey's legal AI technology is not yet reliable enough to handle complex assignments independently, and the company's own benchmark makes that limitation unusually clear.

In May 2026, Harvey introduced its Legal Agent Benchmark to test how well AI systems perform on realistic legal assignments.

Artificial Analysis subsequently evaluated models using 120 private tasks across 24 practice areas, including mergers and acquisitions, taxation, litigation, capital markets, and bankruptcy. We separate what legal AI does reliably today from what is still being tested in our report on legal tech.

These assignments go much further than answering individual legal questions. A model might need to review several documents, identify relevant provisions, apply detailed instructions, and produce a complete legal deliverable.

The researchers used an all-pass score: an assignment counts as successfully completed only when every requirement is satisfied.

In the July 2026 evaluation, the leading configuration, Claude Fable 5 with an Opus 4.8 fallback, achieved 14.2%. The next two models each scored 7.5%, and 13 of the 28 models evaluated failed to complete a single assignment perfectly.

Yet the leading model satisfied 93.6% of individual evaluation criteria.

That's quite a gap. AI can get most of a legal assignment right while still overlooking a detail that prevents the finished document from meeting professional requirements.

We should also be precise about what was tested. Artificial Analysis evaluated underlying AI models on legal tasks designed by Harvey, rather than measuring the full Harvey platform against experienced lawyers. The benchmark has since been updated to version 1.1, whose results are not directly comparable with those initial scores.

Even with that distinction, the initial results give us strong reason to be cautious about claims that legal AI can already handle complicated assignments without supervision.

Result	Initial Harvey LAB-AA evaluation
Legal assignments tested	120
Practice areas	24
AI models evaluated	28
Leading model all-pass score	14.2%
Second-highest score	7.5%
Third-highest score	7.5%
Fourth-highest score	6.7%
Fifth-highest score	5.0%
Models completing no tasks perfectly	13
Leading model's individual-criterion pass rate	93.6%

Is Harvey faster than a lawyer at legal research?

Harvey can make legal research dramatically faster, especially when the assignment involves reading large amounts of material and preparing an initial analysis.

We have compelling evidence that advanced AI tools can improve legal productivity, although the strongest controlled studies do not specifically test Harvey.

A 2026 study led by Daniel Schwarcz examined how upper-level law students performed legal assignments using either Vincent AI, OpenAI's o1-preview, or no AI assistance.

The researchers tested six kinds of legal work and found statistically significant productivity gains ranging from 50% to 130% across five of them.

The largest improvements appeared in analyzing legal complaints, drafting memoranda, and preparing persuasive letters.

A 130% productivity increase means approximately 2.3 times as much quality-adjusted work per unit of time. That is a substantial difference for assignments that normally require hours of reading and writing.

The research also found improvements in legal work quality, making the results more interesting than a simple speed comparison.

There was one exception. Neither AI tool produced a meaningful improvement on the study's nondisclosure agreement task.

We cannot transfer these exact percentages to Harvey or assume experienced practicing lawyers would achieve identical results. But the study helps explain why legal firms are adopting AI so aggressively: the potential productivity gains are already large enough to change how certain assignments are staffed.

Can AI write better contracts than an experienced lawyer?

The best AI tools can already produce contract drafts that are as reliable as those written by experienced lawyers, and some outperform the strongest human participants on individual tests.

A September 2025 study by Legal Benchmarks compared 13 AI tools with in-house commercial lawyers averaging approximately 10 years of experience.

Researchers evaluated 30 contract-drafting tasks, comparing reliability, usefulness, and workflow support.

The results were strikingly close.

Human lawyers produced reliable first drafts in 56.7% of the evaluated tasks. The AI systems averaged 57%.

The strongest human participant achieved 70% reliability, while the leading AI tool reached 73.3%.

AI also had an enormous speed advantage. Human lawyers took an average of 12 minutes and 43 seconds per task, compared with less than one minute for AI-generated drafts.

However, the lawyers scored slightly higher on usefulness, averaging 7.53 compared with 7.25 for AI on a nine-point scale.

The study did not test Harvey directly. Its AI participants included general-purpose tools such as ChatGPT and Claude, alongside specialized legal products.

Still, this is one of the clearest comparisons available between AI and experienced legal professionals performing the same work.

Producing a first contract draft is no longer an exclusively human advantage. Lawyers remain essential for adapting the draft to a client's specific commercial interests, negotiating terms, and approving the final agreement.

Contract drafting measure	Human lawyers	AI tools
Experience of human comparison group	Approximately 10 years	Not applicable
Tasks evaluated	30	30
Systems or participants compared	Human lawyer baseline	13 AI tools
Average first-draft reliability	56.7%	57.0%
Best individual reliability	70.0%	73.3%
Average usefulness score	7.53/9	7.25/9
Average drafting time	12 min 43 sec	Under 1 minute
Human supervision for client delivery	Professional review	Still required

Can Harvey review hundreds of contracts better than junior lawyers?

Harvey has a strong practical advantage over manual junior-lawyer review when a legal team needs to extract the same information from hundreds of contracts.

Imagine a company acquiring a business with 800 commercial agreements.

Its lawyers need to identify change-of-control provisions, termination rights, assignment restrictions, exclusivity clauses, and unusual liability terms.

Traditionally, associates would read the agreements, record their findings in spreadsheets, and prepare summaries for senior lawyers.

Harvey's Review Tables and Vault products are designed to handle large collections of documents and organize this type of information.

The software can apply consistent review instructions across the collection, extract relevant language, and help lawyers focus on provisions that deserve closer examination.

Harvey also allows firms to create customized workflows for recurring legal assignments. Its 2026 product updates added capabilities such as incorporating document review tables into agents and editing Word files directly.

These developments are especially useful for mergers and acquisitions, commercial contract management, and compliance reviews.

However, Harvey has not published a sufficiently rigorous independent comparison establishing that its complete document reviews have fewer errors than reviews conducted by junior associates under normal supervision.

The risk is easy to understand. If an AI system consistently misinterprets a specific kind of contractual provision, the same mistake could appear across hundreds of documents.

For now, Harvey looks particularly valuable for producing the first structured review. Lawyers still need to check the important findings, investigate exceptions, and decide which contractual issues actually matter to the transaction.

Does Harvey make fewer legal mistakes than human lawyers?

There is currently no reliable evidence that Harvey makes fewer serious legal mistakes than experienced attorneys across comparable assignments.

Legal AI can produce convincing answers that contain incorrect conclusions, inaccurate citations, or misleading interpretations of genuine cases. Wrong conclusions and misleading readings of real cases are risks we rate by likelihood in our market report about legal tech.

Researchers at Stanford Law School tested specialized legal research systems from LexisNexis and Thomson Reuters, publishing their findings in the Journal of Empirical Legal Studies in 2025.

The evaluated products hallucinated between 17% and 33% of the time, despite using legal research technologies intended to ground their answers in authoritative materials.

These were not Harvey products, so the results should not be treated as Harvey's error rate.

They do show why access to legal databases does not guarantee an accurate legal answer.

An AI system might find the correct judgment but misunderstand its application, cite a case that has been limited by later decisions, or miss an important exception.

Of course, human lawyers also make mistakes. But proving that Harvey makes fewer errors would require testing both on the same assignments, using independent legal experts to evaluate the results.

We still lack enough published head-to-head evidence of that kind.

Until those comparisons become available, claims that Harvey is more accurate than a qualified lawyer remain unproven.

Is Harvey better than a junior lawyer but worse than a senior lawyer?

Harvey is most competitive with the routine work junior lawyers traditionally perform, while senior attorneys have a much clearer advantage when legal decisions require experience and judgment.

Research, first drafts, and document review occupy substantial amounts of junior associates' time. Harvey can assist with all three.

Senior lawyers are usually responsible for more difficult decisions: determining negotiating priorities, evaluating whether a legal argument is worth pursuing, and advising clients when several options carry different risks.

A fascinating 2026 experiment suggests that professional experience may also determine who benefits most from AI.

David Autor and his colleagues conducted a randomized study involving 133 patent lawyers across 11 American intellectual property firms.

After 10 days of access to an AI drafting assistant, participating lawyers produced work that scored 0.34 standard deviations higher than the control group. After 90 days, the improvement reached 0.38.

But the researchers also tested what happened when the lawyers worked without AI after three months.

Senior lawyers who had used AI performed better than their counterparts in the control group, showing an improvement of 0.45 standard deviations. Junior lawyers showed no average improvement in their independent judgment.

The pattern is especially interesting because junior lawyers had benefited substantially while using AI.

More experienced lawyers appeared better able to turn AI assistance into lasting improvements in their own skills.

Harvey may eventually make senior lawyers even more valuable relative to junior associates, particularly when experienced attorneys can combine their judgment with much faster research and drafting.

Can Harvey replace the work of a junior associate?

Harvey can already replace portions of a junior associate's workload, but there is little evidence that law firms can eliminate most junior lawyers without creating new problems.

The first consequence of AI adoption is likely to be a reduction in the hours required for repetitive tasks. Hours AI saves on repetitive legal work shape market size, which we estimate from the ground up in our legal tech market report.

A due diligence project that once required several associates to review documents may increasingly be completed by a smaller team using AI-assisted workflows.

But junior associates also need those assignments to develop practical experience.

This creates a problem that firms cannot solve simply by buying better software.

The 2026 Thomson Reuters Future of Professionals Report found that 71% of surveyed professionals believed early-career employees needed structured support from experienced colleagues to develop skills that AI might displace.

Legal professionals also expected the time required to develop trusted independent judgment to increase by approximately 1.7 years.

These findings reflect professional expectations, rather than measured long-term career outcomes.

Nevertheless, they highlight a problem already supported by the patent-lawyer experiment: using AI to produce better work does not necessarily make an inexperienced lawyer better at thinking independently.

Law firms may therefore need fewer junior hours for document production while investing more heavily in training and supervision.

The question for firms is becoming quite practical. If AI takes over much of the work that traditionally trained junior lawyers, how will the next generation of senior lawyers acquire the necessary experience?

Can Harvey give legal advice without a real lawyer?

Harvey can explain legal issues and generate proposed advice, but clients still need a qualified professional for regulated legal representation and decisions carrying significant legal consequences.

In the United States, professional legal practice is governed by state rules concerning competence, confidentiality, unauthorized practice, and responsibility for client matters.

The American Bar Association addressed generative AI directly in Formal Opinion 512, published in July 2024.

The opinion explained that lawyers using these technologies must still meet their professional duties, including verifying outputs when necessary, protecting confidential information, and communicating appropriately with clients.

A lawyer cannot avoid responsibility for a faulty recommendation simply because the advice originated from an AI tool.

The same basic issue appears in other jurisdictions, although the rules vary.

A question about American securities law, English commercial litigation, or French employment law may require different professional qualifications and a detailed understanding of local legal procedures.

Harvey can help prepare the analysis, organize supporting authorities, and draft an initial recommendation.

But a client facing a serious legal dispute or transaction still benefits from having an identifiable professional who understands the situation, exercises judgment, and accepts responsibility for the advice.

Could Harvey handle a complicated lawsuit by itself?

Harvey cannot currently manage a complicated lawsuit independently, although it can help lawyers complete several time-consuming parts of litigation.

Litigation involves much more than preparing written arguments.

Lawyers must determine which facts are credible, evaluate witnesses, respond to new evidence, negotiate with opposing counsel, and decide whether a settlement is preferable to continuing the case.

Harvey can assist with reviewing evidence, summarizing depositions, finding relevant legal authorities, and preparing draft submissions.

These capabilities become particularly valuable in large commercial disputes, where lawyers may need to examine thousands of documents.

But the strategic decisions are harder to automate.

Consider a company accused of breaching a contract. An AI system might identify several potentially useful defenses, yet the lawyer must decide whether raising those arguments could expose damaging information or weaken settlement negotiations.

The same problem arises when assessing an unreliable witness or interpreting how a judge is likely to respond to a particular position.

Complex litigation also changes as new evidence emerges. A strategy that looked sensible at the beginning may become counterproductive several months later.

Current legal AI systems have not demonstrated the ability to handle that entire process reliably.

Harvey can make litigation teams more efficient today. Replacing the lawyer responsible for the case remains a much more distant prospect.

Is Harvey actually making lawyers better at their jobs?

AI tools are already helping lawyers produce higher-quality work, and some of the strongest evidence comes from controlled experiments rather than company marketing.

The patent-lawyer study led by David Autor is particularly useful here.

All 133 participating lawyers worked on simulated patent-drafting assignments, and independent patent attorneys evaluated the results without knowing which participants had received AI assistance.

After 10 days, access to AI improved average drafting quality by 0.34 standard deviations, corresponding to an increase of approximately 10 percentile points in the control group's score distribution.

After 90 days, the improvement reached 0.38 standard deviations, or roughly 11 percentile points.

The gains appeared consistently across the evaluated dimensions of quality.

However, the researchers found something revealing about the distribution of results.

AI reduced the frequency of weak drafts and increased the frequency of good ones, without producing a noticeable increase in exceptional drafts.

AI helped lawyers avoid mediocre work more effectively than it helped them produce outstanding work.

That is still commercially important.

A law firm producing consistently good documents with fewer revisions can serve clients faster and spend more time on complicated decisions.

We should be careful about applying the experiment directly to Harvey, since it used a custom AI patent-drafting assistant. Yet the findings establish that legal AI's benefits can extend beyond saving time.

Is Harvey better than ChatGPT or other legal AI tools?

Harvey offers a more complete environment for professional legal workflows than a general-purpose chatbot, but there is no clear evidence that its underlying legal answers are consistently better than those from ChatGPT or Claude.

Harvey competes with several different types of products.

OpenAI and Anthropic develop general-purpose AI models capable of legal reasoning and document analysis. LexisNexis and Thomson Reuters offer established legal research systems. Specialized competitors such as Legora focus on legal workflows and enterprise customers. Who competes with whom across general AI models, research tools and workflow specialists is shown on the market grids in our report covering the legal tech market.

Harvey's advantage comes largely from how these capabilities are packaged for professional use.

Lawyers can work with document collections, structured review tables, knowledge sources, and customized agents instead of repeatedly uploading files and rebuilding instructions.

There is evidence that workflow design makes a substantial difference.

The Legal Benchmarks contract study found no meaningful overall difference in output reliability or usefulness between specialized legal AI products and general-purpose systems.

The researchers identified integration with existing tools, including Microsoft Word, as an important way specialized platforms differentiated themselves.

Harvey has also started developing more of its own AI infrastructure. In 2026, it introduced Harvey Tenet, a post-trained open-weight model intended to improve its control over legal AI capabilities.

This makes Harvey's strategy particularly interesting.

The company does not need to build the world's best general-purpose language model to win customers. It needs to provide a more useful environment for lawyers handling real documents, recurring assignments, and sensitive client information.

Whether that advantage lasts will depend on how quickly general-purpose AI providers and established legal software companies improve their own professional workflows.

Is Harvey growing fast enough to change the legal industry?

Harvey is growing at an extraordinary rate, and the scale of its revenue and customer expansion suggests that legal AI has become a serious commercial market.

According to Sacra, Harvey reached approximately $50 million in annual recurring revenue at the end of 2024.

That figure increased to $195 million by the end of 2025, representing 3.9 times the earlier level.

Sacra estimates that Harvey crossed $400 million in annual recurring revenue by August 2026.

Across those observations, Harvey's annual recurring revenue increased more than eightfold in roughly 20 months.

Its valuation climbed rapidly too, reaching $8 billion in December 2025, $11 billion in March 2026, and $15.5 billion in September.

The company nearly doubled its valuation in nine months.

We should treat Sacra's revenue numbers as third-party estimates rather than audited company disclosures. Annual recurring revenue is also a subscription run rate, not the same as revenue recognized over a completed financial year.

Still, the combination of rapidly expanding subscription revenue, growing usage within existing accounts, and substantial venture financing tells us more than any isolated announcement.

Harvey has reached a scale where major customers are building workflows around its technology.

That raises the stakes for law firms. If one organization learns to complete routine assignments with much less labor, competitors may eventually need similar tools to remain price-competitive.

But commercial growth measures demand for Harvey, not whether the software has become a better lawyer than the people using it.

Will Harvey make lawyers cheaper for clients?

Harvey should help bring down the cost of repetitive legal work, but law firms have been surprisingly slow to change how they charge clients.

The problem is straightforward.

If AI reduces a research assignment from five billable hours to two, the client may expect to pay less. A law firm depending on hourly billing, however, may lose revenue from the efficiency improvement.

The 2026 Thomson Reuters Future of Professionals research shows that clients are increasingly aware of this tension.

Among surveyed in-house legal professionals, 71% expected outside firms to change their commercial models as AI use increases.

Yet 62% of law firm professionals reported that their firms had not changed their pricing structures in response to AI.

Another finding shows how quickly expectations are changing. According to the broader Thomson Reuters research, 78% of corporate clients considered AI-enabled quality improvements very important or essential, while only 6% believed most or all of their providers were delivering them.

As seen above, the technology is already capable of improving legal productivity substantially. The unanswered question is how much of that improvement reaches the customer.

There are several plausible outcomes.

Firms could charge less for standard contracts and document review. They might offer fixed-price packages for work previously billed by the hour. Or they could maintain fees while delivering faster turnaround and more detailed analysis.

For highly specialized legal advice, clients may continue paying premium prices because the lawyer's judgment remains the scarce resource.

Routine document work faces much greater pricing pressure.

Are corporate clients starting to prefer Harvey over traditional law firms?

Corporate clients are increasingly demanding AI-powered legal services, but there is little evidence that they want to abandon human lawyers and rely entirely on Harvey.

The more immediate change concerns how companies choose their outside counsel.

In the 2026 Thomson Reuters research, 32% of in-house legal professionals said they were already reconsidering, or expected to reconsider within 12 months, relationships with firms that failed to demonstrate clear AI-enabled value. Clients asking their firms to prove AI value is one of the pressures that can speed the market up, which we weigh in our market report about legal tech.

Among respondents reconsidering providers, approximately one-third estimated that more than $1 million in annual work could be affected.

That is a meaningful commercial threat for law firms that are slow to adopt new technology.

The pressure is also coming from employees. The same legal industry research found that 24% of law firm professionals would reject a job offer if the employer did not provide access to professional-grade AI tools.

Meanwhile, 34% reported using AI tools their organization had not officially approved.

Together, these findings suggest AI adoption is moving faster than some firms' internal policies.

Clients want better service, employees want tools that help them work efficiently, and firms face pressure to modernize without exposing confidential information.

Harvey benefits from all three developments.

The companies buying legal services are increasingly interested in what a law firm can accomplish with AI, rather than whether it employs the largest possible team of associates.

Is Harvey better than a lawyer today?

No. Harvey is currently better than human lawyers at several narrow, repetitive legal tasks, but experienced attorneys still outperform it when a matter requires complete professional judgment and accountability.

The strongest evidence comes from comparing actual legal tasks.

In a controlled contract-drafting study, leading AI tools produced first drafts that matched or slightly exceeded experienced lawyers' reliability while taking less than a minute, compared with almost 13 minutes for humans.

Other experiments found productivity improvements of 50% to 130% on selected legal assignments, with measurable gains in work quality.

That is already enough to change how legal research, document review, and drafting are performed.

However, the available evidence becomes much weaker when we move from individual tasks to complete legal assignments.

Harvey's initial Legal Agent Benchmark showed that even leading AI models struggled to satisfy every requirement of complex professional work. Independent studies have also found substantial accuracy problems in specialized legal research systems.

Experienced lawyers remain particularly valuable when interpreting uncertain facts, negotiating difficult agreements, developing litigation strategies, and advising clients on decisions with serious legal or financial consequences.

The commercial implications are already visible. Harvey has become a major technology provider to leading law firms, while clients are beginning to demand the productivity benefits of AI.

We therefore expect the biggest disruption to affect how many hours lawyers spend on each assignment, how firms train junior associates, and what clients are willing to pay for routine work.

For a simple first draft or large-scale document review, Harvey may already be the better starting point.

For an important contract negotiation, complex lawsuit, or difficult legal decision, an experienced lawyer remains the stronger choice.

And for most professional legal work, the combination of an experienced lawyer and capable AI software is currently more compelling than either working alone.