Will AI Replace Lawyers? What the Evidence Says in 2026

Richard Susskind refuses to answer the question “what is the future of lawyers?” because, he says, the question assumes lawyers have a future. At a June 2026 Axiom fireside in London he added the line that stings: lawyers are the slowest adopters of technology of any profession “except the clergy”.

Will AI replace lawyers? is the question every associate, student, client and journalist asks, and almost everyone who answers it is selling something: a platform, a course, a doom narrative or a reassurance. What follows is the evidence instead: two randomised trials, the hiring and billing numbers in both directions, the AI-native firms and what they charge, the pyramid thesis, and the commentators whose predictions have held up. The answer at the end is not a prediction; it is a list of what to do in the next twelve months.

The question behind the question

The number that will not die is Goldman Sachs’ March 2023 estimate of “particularly high exposures in administrative (46%) and legal (44%) professions”. Goldman’s own August 2025 update, as computed by Artificial Lawyer, implies roughly 228,000 of 1,322,000 US legal jobs, 17.2%, are exposed, and the same article makes the distinction most coverage skips: “being exposed to the risk of job replacement by AI, and it actually happening are two different things entirely.”

Exposure is a statement about tasks. Clio’s 2024 Legal Trends Report estimated that 74% of hourly billable tasks are potentially automatable, a different claim from 74% of lawyers being replaceable. On the outer limit the profession is clear: in Thomson Reuters’ 2026 survey, 96% said AI representing clients in court would be “a step too far”.

So the honest form of the question is the one Thomson Reuters’ chief executive Steve Hasker put in the 2025 Future of Professionals report: “AI will not replace professionals, but AI-powered professionals will.”

What the randomised trials show: faster first, then better

The randomised controlled trials of AI assistance in legal work, both with Minnesota’s Daniel Schwarcz among the authors and the second run jointly with Michigan, come in two acts.

Act one, GPT-4, 2023-24. Choi, Monahan and Schwarcz gave 60 University of Minnesota law students four realistic tasks, with or without GPT-4, and had the work blind-graded. Their finding, in 109 Minn. L. Rev. 147: “access to GPT-4 only slightly and inconsistently improved the quality of participants’ legal analysis but induced large and consistent increases in speed.” The lowest-skilled participants gained most, which led to a line worth remembering: “Because AI tools have an equalizing effect on performance, they may also promote equality in a famously unequal profession.”

Act two, reasoning models and retrieval, 2025. Schwarcz, Manning, Prescott, Barry, Cleveland and Rich ran a second trial with upper-level students using OpenAI’s o1-preview, vLex’s Vincent AI (a retrieval tool) or no AI on six tasks. The result: “both AI tools significantly enhance legal work quality, a marked contrast with previous research examining older large language models like GPT-4”, with productivity gains in five of six tasks “of anywhere from 50% to 130%”. The study was funded by two law firms and the two law schools, not by OpenAI or vLex.

One vendor benchmark deserves a footnote: in Vals AI’s February 2025 study, Harvey scored 94.8% on document question-answering against 70.1% for lawyers, yet lawyers beat every tool on redlining, 79.7% to 65.0%. The humans are still better at knowing which change matters.

What the market shows: hiring, hours, rates

If AI were replacing lawyers you would see it in hours billed. You do not. Citi’s mid-year data, reported by Best Law Firms in September 2026, shows industry billable hours up 4.2% in the first half of 2026 against a historical norm of 1.5-2%, with revenue up 11.7%. Bloomberg’s reading: “With billable hours surging even as firms adopt AI, the industry is not seeing the technology eat into demand.”

The rate side is stranger. Roughly 90% of legal dollars still flow through hourly arrangements and Am Law 100 profits per lawyer are up 53.7% since 2019 (Thomson Reuters’ 2026 State of the US Legal Market); associate rates have risen 33% since 2023 to an average of $798, according to Persuit data cited by the FT, which is why, as the FT reported in September 2026, Goldman Sachs, Morgan Stanley and Citi are pressing firms to cut fees. The pricing fight is in AI and the billable hour.

Hiring is where the question bites, and the data runs both ways.

There is also a demand shock nobody predicted: lawyers on Reddit report that pro se filings have “doubled or tripled” because the other side has AI too. The sentiment is in what lawyers really think about AI and the numbers in legal AI statistics.

The AI-native firms: Garfield, Crosby, Eudia, Lawhive, Norm, Covenant

The strongest evidence that something is being replaced comes from firms built to do without it. The AI-Native Law Firm report’s test: “in a bolted-on firm, removing the AI tools is an inconvenience; in an AI-native firm, it breaks the operation.”

Source: IBA; Crosby; AI-Native Law Firm report.

Garfield’s regulator-approved design forbids the AI from proposing case law, which tells you where the profession’s supervisors think the danger is. And Crosby’s founders framed their target in one sentence: “America’s top 100 law firms made a combined $69 billion in profit last year, greater than Google’s R&D budget. Every cent was paid out to the firms’ partners as compensation.” The model is examined in AI-native law firms explained.

The incumbents’ response: Kirkland’s $500 million and the pyramid becoming a cylinder

Incumbents are spending, though less than the headlines suggest. Kirkland & Ellis committed $500 million over three to four years to a proprietary platform, about $100 million in 2026. Tom Martin’s arithmetic in the same piece is the corrective: about 1% of the firm’s $10.6 billion revenue, against 13% reinvested in R&D by software firms, so “a 1% commitment reads less like a moonshot and more like minimal maintenance”.

The rest of Big Law has bought rather than built. Latham has 3,600-plus attorneys on Harvey; Freshfields deployed Claude to 5,700 employees and saw usage grow about 500% in six weeks; Harvey itself, valued at $15.5 billion after its 9 September 2026 round, says 80% of the Am Law 100 use it.

So far these deployments change the associate-to-partner ratio, not headcount. Harvey’s June 2026 guidance says a first-year who spent 40 hours marking up a services agreement “might now spend 5 hours reviewing an agent’s markup”, and warns that craft dies without “intentional friction”: drafting by hand before comparing with the agent.

The most quoted structural prediction is Citi and Hildebrandt’s December 2025 client advisory, as reported by Reuters: 86% of large firms plan to grow associate ranks through 2027, but only 35% plan larger first-year classes, and 63% expect generative AI to change the ratio of associates to partners by 2035, “from a pyramid to a cylinder”. Citi’s Gretta Rusanow: “If we’re going to move away from the pyramid model, because tens of thousands of hours are knocked out through generative AI… how do you grow those individuals into those strategic advisors?”

Thomson Reuters’ 2026 Future of Professionals report found 48% fear a negative effect on the development of independent judgement, and legal professionals expect the path to “trusted judgment” to lengthen by 1.7 years. LexisNexis’s Mentorship Gap study of nearly 900 UK lawyers found 72% name deep legal reasoning as the biggest skills gap and only 2% believe AI strengthens learning. An anonymous senior associate, quoted on the AI and the Future of Law podcast: “For people who already have domain knowledge… AI is an enhancement. And for those who don’t, it’s clearly a replacement.” The junior-specific evidence is in will AI replace junior lawyers.

Susskind: the clergy, thirty minutes a day, the fence at the top of the cliff

Susskind’s June 2026 fireside is the most useful single framing of the decade. Short term, AI is an efficiency tool for two to three years; by the 2030s it “will be a tool that empowers non-lawyers”, and the profession is “sleepwalking” towards a world of near-general intelligence. “These are different conversations, and they need to be happening in different rooms.”

Three points survive contact with a law firm. The adoption recipe: a general counsel who made the whole team use AI 30 minutes a day for six months found the team “was transformed”, whereas pilot groups create “two classes of lawyer”. The diagnosis of resistance: “irrational rejectionism”. And the picture of what agents are for: AI roaming an organisation’s data continuously is a “fence at the top of the cliff” rather than an “ambulance at the bottom”. The future lawyer will “build, maintain, supervise, and improve the systems”.

Build me a six-week plan of 30 minutes a day to become competent with [Claude / ChatGPT / Copilot] for [transactional / litigation / in-house] work. Each day: one task on my own anonymised materials, the prompt pattern it teaches, a success criterion I can check, and what could go wrong. Week 1: low-stakes personal tasks. Week 2: summarising and extraction with page references. Week 3: drafting from a verified source, with a verification step every day. Week 4: review against a playbook. Week 5: a reusable Project with my standing instructions. Week 6: one full workflow end to end. Never include a task that asks for case law without supplied sources. End with a self-assessment I can repeat in six months.

What gets replaced: tasks, not roles

The replacement that is actually happening shows in studies that compare humans and models on a single task. Onit’s “Better Call GPT” study found GPT-4 matched junior lawyers on identifying contract issues (F-score 0.871 against 0.860) but was weaker at locating them (0.686 against 0.770 for legal-process outsourcers), at $1.24 per contract for GPT-4 and $0.02 for Claude against $74.26 for a junior. Thomson Reuters’ Legalweek 2026 coverage put it in one line: “The billable hour was never the thing in danger, rather it’s the person billing the hours. It’s the associate.”

The redefinition is being designed on purpose at some firms. Ropes & Gray’s TrAIlblazers track lets first-year associates spend 20% of their creditable time learning and innovating with AI; Sidley Austin’s London office added a mandatory AI Knowledge Lab seat to every training contract in September 2026 (“The Lab is a working seat, not a classroom”). The skills involved are in AI skills lawyers need, and the agents doing the first pass in what is an AI agent for lawyers. Two exercises make it concrete; first, an honest audit of a week.

Here is everything I did last week, with rough hours: <week>[paste anonymised time entries]</week>. Classify each item as AUTOMATE (a fixed standard exists and the output can be checked quickly), ASSIST (a model can produce a first draft or extraction that I must read in full), JUDGEMENT (client relationship, strategy, negotiation, advice under uncertainty, anything I sign) or UNSURE. For each AUTOMATE and ASSIST item, name the input I would have to supply and the check I would run. Total the hours per category. Then tell me which two ASSIST tasks to try first, and which JUDGEMENT tasks I risk under-investing in if the others get faster.

The second is Harvey’s “intentional friction”, turned into a supervised exercise for a junior.

Set up an exercise for a first-year associate on [reviewing a services agreement markup]. Step 1: the associate does clauses [8-12] by hand in [60] minutes, no AI. Step 2: run our standard review prompt on the same clauses. Step 3: a comparison table: issues found by the associate only | by the AI only | by both | by neither (the partner fills this column). Step 4: five written questions on why the AI missed what it missed and why the associate missed what they missed. Step 5: the associate writes the final markup. Produce the instructions, the table template and the questions. The point is that the associate learns to check, not to accept.

Shapiro: not using the tools is becoming the harder position to defend

The most-read practitioner account of 2026 came from Zack Shapiro of Rains LLP, a two-person firm, who published “How I Actually Practice Law with AI in 2026” on X on 27 February 2026; Artificial Lawyer put its reach at over seven million views. His claim: a general-purpose model with custom “skills” encoding one lawyer’s judgement beats vertical legal AI for a small firm, and “The entire gap between ‘AI is a toy’ and ‘AI changed my practice’ lives in the quality of your instructions.”

His answer to the replacement question is the one this page endorses: “If you’ve spent 10 or 20 years developing judgment in your practice area, you are sitting on exactly the asset that AI makes more valuable, not less.” And his warning inverts the debate of 2023: “We are approaching the point where not using these tools is the harder professional responsibility position to defend.”

The regulators are moving the same way. ABA Formal Opinion 512 said in 2024 that “it is conceivable that lawyers will eventually have to use them to competently complete certain tasks for clients.” The UK Jurisdiction Taskforce’s July 2026 legal statement went further, at paragraph 67: “It is important to be aware of the possibility that a professional could also be liable for failing to use AI for a task when a professional exercising reasonable care and skill would have done so.”

MacEwen and Furlong on the business model and the pipeline

Bruce MacEwen of Adam Smith, Esq. is the clearest voice that the market has already changed. In December 2025: “I cannot see how the billable hour revenue model survives the arrival of gen AI.” In July 2026 he described AI as “a brand new third legal service provider” alongside firms and in-house teams: “This is not a cycle; it’s never going back.” A Ford and Microsoft survey he cites found 74% of firms rate themselves ahead of clients on AI while only 30% think firms meet client expectations: “the math doesn’t work”.

Jordan Furlong’s concern is the pipeline. His April 2024 analysis of Thomson Reuters and NALP data showed entry-level private-practice jobs falling from 20,611 in 2007 to 16,390 in 2017 and Am Law 100 first-year hiring down nearly 17% in one cycle, before generative AI was widely used. His warning: “we will effectively be licensing unemployable lawyers.” And to firms hiring Generation Z: “be careful not to confuse ‘digitally native’ with ‘technologically savvy’”.

Crosby’s Ryan Daniel names the skill both imply: “Explaining things is something that… is going to be a very prized skill for not just lawyers, but for any domain experts, but in particular lawyers.”

What to do in the next twelve months

The evidence supports five actions.

1. Use the tools every working day. Susskind’s 30 minutes for six months, the whole team, not a pilot group. LexisNexis’s September 2026 survey found 46% of US lawyers believe their career will suffer if their organisation does not embrace AI, up from 28% a year earlier.
2. Learn to write instructions, not prompts. Shapiro’s point that the gap “lives in the quality of your instructions” is the thesis of prompt engineering for lawyers.
3. Verify like the trials tell you to. Eleven hallucinated citations from the model that most improved quality. Every citation opened, every quotation checked, every number re-done, your name still on the review.
4. Encode your judgement. Turn the way you review, negotiate and advise into playbooks and standing instructions; that is the asset Shapiro describes.

Help me write down how I do [reviewing a supplier-side SaaS agreement]. Interview me, ten questions at a time, up to forty: what I look at first and why; the positions I always take and the ones I trade; the mistakes juniors make that I catch; the facts that change my answer; how I know when I am done. Do not draft until the interview ends. Then write the standing instructions in my words: safety rules first (never invent authority, tag every citation [VERIFY], say "NOT IN DOCUMENT" rather than infer), then the method as numbered steps, then the judgement calls, then what must be escalated to me. Under 700 words. List the anonymised documents I should attach as examples.

5. Pick the training format that works. Paul Weiss found its first PowerPoint session “ineffective” and moved to a hands-on prompting workshop; the options are compared in AI training for lawyers.

Where to go next: the careers and future hub collects the junior-lawyer, skills and AI-native-firm guides; the pricing side is in the billable-hour guide linked above. Susskind’s thirty minutes a day is the premise of AI Lab for Lawyers: four live two-hour sessions, browser tools only, on the workflows that make you the lawyer who uses the tools well.