No, not on the evidence gathered so far, but the newest lawyers already feel it. The US Bureau of Labor Statistics still forecasts 4% growth in lawyer jobs from 2024 to 2034: about 31,500 openings each year, at a median wage of $151,160 in May 2024. Those numbers describe a profession that keeps hiring. They say nothing about which rung of it AI is emptying first.
One global law firm has halved its intern class, from 1,000 trainees to 500, according to interviews conducted by Harvard Law School’s Center on the Legal Profession. The centre spoke to 50 chief legal officers, law firm partners, legal operations leaders and corporate executives for a report published in July 2026. The cuts it found sit at the bottom of the profession, not the top.
more
Where the software already does the work
The tasks most exposed to automation share three features: bulk text, repeatable steps, and patterns a model can learn to spot. Contract review, e-discovery, due diligence, compliance work and the first pass at legal research all fit that description. Vendors have built specifically for them: Relativity’s aiR handles document review in litigation, LexisNexis sells research and drafting tools, and Thomson Reuters has folded its CoCounsel product into the same pitch.
Adoption has moved fast. The UK’s Solicitors Regulation Authority found three-quarters of the largest solicitors’ firms using AI by the end of 2022, almost double the share it had recorded three years earlier. A third of mid-sized firms were trying AI out too, the regulator found, and the share was higher still among large firms: at least three in five.
Who is actually losing work
The Harvard interviews put numbers on what that adoption costs junior staff. A chief legal officer at a Fortune 500 logistics company said he had cut $1 million from his external counsel budget by building AI capability in-house. A chief operating officer for legal at another company said the chief financial officer now expects legal spend to fall by 30%. A third interviewee, a chief legal officer at a large financial institution, already lists five AI “agents” on his organisation chart, and expects to double that count within the year.
Mary O’Carroll, a legal operations leader, said in May 2026 that in-house teams increasingly test AI on a matter first and only call an outside lawyer if the work still needs one. “We get to 80 percent on our own,” is how one team described it to her.
A partner at a mid-size New York firm drew the line where the numbers suggest it falls: the tools cannot do his job, but they can handle “aspects of a junior associate’s”. The report adds that clients no longer want to pay a person for that work.
What the software still cannot do
Ask the lawyers doing the work day to day, and the limit shows up quickly. Lucas Hale, a junior associate at the law firm McDermott Will & Schulte, uses AI to sift long documents and draft citations. He says it fails on any question a court has never ruled on before, calling that “the kind of thinking that the tool can’t do”. Allison Douglis, a senior associate at Jenner & Block, uses a language model to start her research, not to finish an argument. “I would much rather work with a junior associate,” she said, than an AI tool.
A benchmark backs up what they describe. ScaleAI’s Professional Reasoning Benchmark, published in November 2025, tested leading models against legal and financial tasks set by professionals in the field. The best model scored just 37% on the hardest legal problems. Afra Feyza Akyurek, the benchmark’s lead author, said the tools are nowhere near acting as a substitute for a lawyer. She added that legal reasoning in large language models still lags behind what people assume it can do.
Joseph Yockey holds the F. Arnold Daum Chair in Corporate Law at the University of Iowa. He frames the gap as a skill new lawyers now need: knowing the point at which a person has to take the decision back. From here on, he argues, competence with AI counts as professional competence.
When the software is confidently wrong
The failure mode with the worst record is invention. Two lawyers in the United States were fined after filing court citations that ChatGPT had fabricated, though the account reporting the case names neither the court nor the date. Courts and bar groups keep repeating the same instruction: a lawyer has to check what a model produces, because these tools invent facts, drop context and sound certain while wrong.
Jobs are already going. The law firm Clifford Chance cut 10% of its London staff in November 2025 and named heavier AI use as a reason. By 2024, 47.8% of attorneys at firms of 500 or more lawyers were using AI tools, according to the American Bar Association. Rudi Miller, a junior associate at Orrick, chose litigation partly because she hoped the courtroom would stay a human stage the longest. As she put it, judges have not yet let a ChatGPT-enabled robot argue a case.
The number that actually tells you something
Total lawyer headcount will keep climbing, on the Bureau of Labor Statistics’ own projection. The figure worth tracking instead is the size of each year’s trainee intake, because the Harvard interviews found displacement starting at the bottom of the pyramid. A profession that keeps its partners while it stops training replacements has not been replaced. It has stopped growing the next generation of the people who would be.
