Lawyers adopted AI months ago.
Their firms are still catching up.

Lawyers made the AI decision. Firms have not caught up.

Individual lawyer use of generative AI has more than doubled in the past twelve months, from 31 per cent in 2025 to 69 per cent now, according to new research from legal AI tracker 8am. Twenty-eight per cent of lawyers now use AI daily, and a further 31 per cent use it several times a week. That is not early adoption. That is most of the profession, working a certain way, on their own initiative.

Firms have not kept pace:

• Only 46 per cent have formally adopted general purpose AI tools.
• Just 34 per cent have adopted legal specific platforms, rising to 58 per cent among firms with 20 or more lawyers.
• Fifty-four per cent provide no AI training at all, and have no plans to introduce any.
• Only 9 per cent have a written AI policy that is actively enforced.
• Forty-three per cent have no formal policy whatsoever, and no plans to write one.

Put simply, most lawyers are already using AI, on tools of their own choosing, with no training, no policy and no firm-level visibility into what they are doing with it. That is not a rollout problem. It is closer to a shadow practice, and it is happening inside firms that still believe they are deciding whether to adopt AI at all.

It is worth being specific about where that shadow practice is concentrated. The tools lawyers have adopted fastest sit squarely in the highest volume parts of the job: first pass contract review, early-stage due diligence, regulatory research summaries. These are exactly the tasks used to justify a junior lawyer's billing line. That is not a coincidence.

The billable hour, not the technology, is the real obstacle

New research from legal education group BARBRI, published in early August, gets at why the governance gap persists. Interviewing leaders across nine firms, from global Am Law 100 practices to newer alternative model firms, the report found that no firm interviewed had built an AI competency framework for its own associate pipeline needs. Firms graded their own rollouts a C. Most tellingly, firms can tell you who has activated an AI tool. Almost none can tell you who has actually changed the way they work because of it.

The report's conclusion is blunt: the billable hour remains the most stubborn barrier to AI adoption, and training alone cannot fix it. Firms are not failing to adopt AI because the technology is not ready. They are stalling because nobody has redesigned the economic model that AI breaks.

Here is the mechanism. A first-year associate billing at roughly 600 dollars an hour for drafting and research made economic sense when that work took hours. Recent research finds 58 per cent of lawyers using AI tools report producing work faster, rising to 65 per cent among those using premium platforms. When that same drafting and research takes minutes, the economic logic for billing at all starts to disappear, not because the work has stopped mattering, but because nobody can justify the invoice line.

The same research flags what firms lose if they solve the pricing problem carelessly:

• Seventy-two per cent cite deep legal reasoning and argumentation as the largest skill gap now showing up in junior lawyers.
• Sixty-nine per cent cite weak source verification.
• A separate UC Berkeley study found generative AI increases total workload rather than reducing it, intensifying burnout rather than relieving it.

Speed is not the same as judgment, and firms that measure AI success purely in hours saved are optimising for the wrong outcome.

Three markets, three ways of not having the conversation out loud

The United States is dealing with this through headcount. McDermott Will and Schulte, formed by last year's merger with Schulte Roth and Zabel, let go of a small number of associates in May, concentrated in New York. The firm attributed the move to aligning with shifting client needs and disciplined decisions about how it invests and grows. To be clear, the firm has not linked the decision to AI, and neither should this newsletter. What is notable is the timing.

That quiet contraction at the junior end sits oddly against what is happening one rung up. Am Law lateral partner hiring data for the second quarter of 2026 shows 75 per cent of moves were poached directly from other Am Law 200 firms, up from 60 per cent in the first quarter, with litigation accounting for 286 of the moves tracked and corporate a further 172. Kirkland and Ellis, the most aggressive hirer in the market, still lost 43 of its own partners to competitors in the same quarter. Senior talent is being fought over more fiercely than at any point in years, while junior headcount at the same firms is being quietly trimmed. That is a two-speed market.

The United Kingdom is dealing with this through investment and messaging. Slaughter and May announced a firm-wide rollout of Harvey's AI platform across every practice area in April, covering multi-jurisdictional M&A, due diligence, regulatory research and document analysis. Managing partner David Johnson framed the move around protecting the investment the firm makes in its people, describing them as the vital human layer that supervises AI. That is a genuinely different posture to a quiet layoff. It is also, so far, a statement of intent rather than a published training plan.

The Magic Circle firms that locked in a uniform 150,000 pound newly qualified salary this summer, covered in our last issue, are making a clear bet that pay retains talent. This month's data suggests pay is necessary but not sufficient. What retains talent is whether the firm can show associates a credible path to the judgment work that used to come from doing the drafting themselves.

The UAE is dealing with this through the market, more openly than the other two. New data from recruiter Beacon Legal, reported by AGBI, shows newly qualified lawyers at local Dubai firms earning around 28,000 dirhams a month, roughly 7,600 dollars, against 63,000 dirhams, around 17,150 dollars, at US firms in the same city. That gap has been growing, not narrowing. Maisa Maarouf, head of HR at BSA Law, put it plainly: AI's impact has been on tasks rather than roles, and firms now need to rethink how they develop talent.

The DIFC Courts, meanwhile, have gone further than either the UK or US regulators, publishing formal guidance requiring lawyers to disclose AI use, independently verify AI generated content, and obtain client consent before deploying it. Junior lawyers in Dubai are now formally tasked with educating their own clients on AI's risks and benefits, an unusual reversal in which the most junior people in the room are expected to be the most fluent in explaining the technology.

What actually happens next

None of this is really an AI story. It is a training story that AI has made urgent. The apprenticeship model in every one of these three markets was funded, quietly, by clients paying for junior lawyers to learn on real matters. AI has not eliminated that funding mechanism outright, but it has made the least defensible part of it, hours billed for work a machine now does faster, much harder to justify.

Firms that respond by simply cutting that billed time without replacing the training it used to fund will, over two or three years, produce associates who are fast and unreliable in exactly the ways this month's research flags: strong on output, weak on judgment and source verification. Firms that use the freed-up time deliberately, to put junior lawyers in front of clients earlier, to have partners review reasoning rather than drafting, to make verification and judgment an explicitly taught skill, will produce lawyers who are both fast and trustworthy. That second group will have a real hiring advantage within two years.

Watch three things before the next issue:

• Whether any major firm becomes the first to publicly attribute headcount decisions to AI rather than euphemism.
• Whether Magic Circle firms rolling out platforms like Harvey publish anything resembling a competency framework, rather than a procurement announcement.
• Whether the Dubai wage gap between local and international firms widens again next quarter, which would suggest the market, not policy, is already pricing this shift in.

How the three markets actually compare

The UK is currently ahead on investment and furthest behind on stated plans for retraining. Slaughter and May's rollout is real money and real infrastructure, but a platform announcement is not a curriculum, and none of the Magic Circle firms making similar bets have published one.

The US is furthest along in letting the market quietly restructure headcount without naming the cause, which is efficient in the short term and corrosive in the long term, because associates who read layoffs as arbitrary rather than explained tend to make career decisions based on fear rather than fact.

The UAE is the most transparent of the three, both in its regulator's guidance and in how openly its wage data reflects the shift, but it is also the market with the least established infrastructure to retrain junior lawyers at scale if local firms need to compete for the same talent international firms are already paying a premium for.

None of the three markets has yet built what the BARBRI research says is missing everywhere: a real competency framework that tells a junior lawyer, concretely, what they need to be able to do that a model cannot.

What this means for decision makers

For firms, the mistake to avoid is treating this as a technology rollout question. The tools are being adopted by lawyers whether the firm has a policy or not. The actual decision in front of every managing partner right now is how the firm intends to fund and structure judgment training now that its old funding mechanism, billing junior time for work AI can do faster, is eroding. Publishing a written, enforced AI policy is a start; only 9 per cent of firms have one. A competency framework that goes beyond tool licensing is the harder and more important step.

For newly qualified lawyers and associates, the instinct to prove value through speed is understandable and, on this month's evidence, exactly backwards. Fast, AI assisted output is now the baseline expectation, not a differentiator. What differentiates is deep reasoning, argumentation and independent verification, the three things firms report junior lawyers are currently weakest on. Lawyers who can point to those skills explicitly, not just to tool fluency, will be the ones firms compete for.

For lateral and partner level moves, the practical signal is which firms can demonstrate an actual training pipeline behind their AI investment, not just the investment itself. That distinction is currently the best predictor of which platform is a genuine long-term bet, and which is a procurement press release.

AI did not remove the need to train judgment in young lawyers. It removed the excuse firms were using to avoid paying for it directly.

NMG Recruitment advises firms and senior lawyers across the UK, UAE and US legal markets. If your hiring plan has moved faster than your training model, we welcome a confidential discussion.