Legally Artificial — September 25, 2026
Workforce Transformation
How AI is changing the legal workforce: hiring, training, and the people.
The Week to Date
Opening Argument
One argument, drawn from the week to date.
Law firms are fighting for AI talent on two fronts at once, and training is where the fight shows first.
Firm Prospects' new lateral market report found that 46 lawyers left Am Law 200 firms for AI companies in the first half of 2026, 22 of them to Harvey alone. That is a small slice of the 15,988 total moves the report tracked, but it concentrates in the companies whose tools those same firms are buying, and 15 of the 46 took the title "legal engineer," a role most Am Law 200 firms still do not carry on their own org chart.
The visible response is capital, not headcount: Cooley, Kirkland and Morgan & Morgan have committed to their own AI teams and platforms this month, with no matching hiring wave yet. In the same week, Gunderson Dettmer trained its newest partners in sales instead, betting that client relationships are what AI cannot take. Law schools cannot even agree which skill to build first: Columbia leans into AI fluency while Chicago and Berkeley restrict it.
The Implication
Ask which fight your own hiring budget is funding: people who can build AI tools, or people clients will choose over one. This week's items suggest firms are placing both bets without saying so, while training the incoming class for only one.
The Findings
What moved in practice this week, and what each one tells you.
-
Am Law 200 firms lost 46 lawyers to AI companies in the first half of 2026, with Harvey alone drawing 22 of them.
Firm Prospects' H1 2026 lateral market report tracked 15,988 moves into and out of the Am Law 200 between January and June. Of the 46 that landed at AI companies, 40 were associates, 5 were partners, and 1 was counsel; 15 took the title "legal engineer." Harvey drew 22, Anthropic 8, OpenAI 5, and Legora 5.
Why it matters This is not a mass exodus against 15,988 total moves. But the destinations concentrate in four companies building the tools those firms buy, and "legal engineer" as a landing title rather than a firm-side one is worth checking again next report.
-
Cooley, Kirkland and Morgan & Morgan are each building AI teams and proprietary platforms this fall, not only trying to out-bid AI companies for the lawyers leaving.
JD Journal reported September 25 that Cooley opened 14 positions under a new Cooley AI unit, with at least four filled by September 24 and senior AI data engineer postings at $220,000 to $250,000; one firm advertised up to $438,000 for a director of applied AI. Kirkland & Ellis has committed $500 million and Morgan & Morgan at least $1 billion over coming years to their own AI platforms.
Why it matters These are capital commitments, not proof of headcount. Watch whether this spending shows up as in-house technical hiring or stays a budget line.
-
Chicago, Berkeley and Columbia adopted opposite AI rules for law students this fall, splitting the incoming pipeline into differently trained cohorts.
Chicago barred laptops, tablets and phones in first-year core courses and now requires internet-free exams, aiming for what it calls "AI-resilient" teaching. Berkeley's default rule bars AI to "conceptualise, outline, draft, revise, edit or translate" submitted coursework. Columbia takes the opposite approach, permitting AI for research and argument-testing while barring reliance on it for the underlying legal reasoning, and adds AI training from orientation onward. At least a dozen US law schools revised their AI policies over the summer.
Why it matters A class trained under Berkeley's restrictions and one trained under Columbia's will not arrive at their first jobs with the same baseline AI fluency, and neither hiring firms nor a shared bar exam are set up yet to sort for the difference.
-
A Silicon Valley firm put its newest partners through a business-development camp this month built on one message: you are in the business of sales, not just law.
Gunderson Dettmer put nine newly promoted partners through training this month emphasizing client relationships and business development as AI absorbs more routine drafting and research. Corporate partner Ivan Gaviria told the group, "you are not Atticus Finch... you have to accept that you are in the business of sales."
Why it matters If relationship-building is the retraining target, that is a bet that client trust, not legal analysis, is the scarcest input left. Test that bet against your own client conversations before a training budget gets restructured around it.
-
A recruiting-industry report projects the fastest-growing legal job titles through 2030 are all hybrid legal-AI roles, led by 39 percent growth for "AI legal specialist."
JD Journal's September 22 write-up of a LawCrossing report lists AI legal specialist (39 percent), legal technology manager (37 percent), cybersecurity legal counsel (35 percent), AI implementation manager (33 percent) and legal AI trainer (32 percent) as the fastest-growing categories. Methodology and sample size were not disclosed.
Why it matters This is a projection from a site with its own interest in the roles it lists, not a labor-market census. Check any of these titles against real job postings and BLS data before a hiring plan gets built on them.
The Practice
This week's deep read on AI in practice.
The war for legal AI talent is running in two directions at once.
Four items this week describe four decisions: who is leaving, who is spending, who is training, and who is projecting growth. Side by side, they describe one unresolved question: is the scarce resource technical AI capability, or the human judgment AI has not reached?
The outbound number is small, and concentrated
Firm Prospects' count of 46 Am Law 200 lawyers who moved to AI companies in H1 2026 is a rounding error against the 15,988 total moves its report tracked. But look at where they went, and what they were called when they got there.
46
Am Law 200 lawyers who moved to AI companies in H1 2026, by destination.SOURCE · FIRM PROSPECTS, H1 2026 LATERAL MARKET REPORT
On the Methodology
Firm Prospects identified these moves from its own database of 15,988 recorded Am Law 200 moves between January and June 2026, not a survey, and the firm-by-firm breakdown has not been independently verified.
Forty of the 46 were associates, not partners with books of business to carry out the door. That is the group whose training a firm has already paid for, and the group law schools are now training under rules that no longer agree with each other.
The response, so far, is money
Cooley opened 14 positions under a new Cooley AI unit this month, with at least four filled by September 24 and senior AI data engineer postings listed at $220,000 to $250,000; one firm advertised up to $438,000 for a director of applied AI. Kirkland & Ellis has committed $500 million and Morgan & Morgan at least $1 billion over coming years to their own AI platforms.
None of those figures is a headcount number. A platform budget still has to be spent on people who can build it, and the data above says some of the people firms need are choosing to build those platforms for someone else instead.
Retraining is where the headcount decision is actually made
Gunderson Dettmer put nine newly promoted partners through a business-development camp this month, telling them, in corporate partner Ivan Gaviria's words, that they are "in the business of sales." That is a bet on a different scarce resource than Cooley and Kirkland are funding: not AI capability, but the client relationship AI cannot yet form on its own.
The incoming class will not arrive trained for either bet uniformly. Chicago and Berkeley have restricted AI use in first-year coursework this fall to protect independent legal reasoning, while Columbia teaches AI fluency from orientation onward. A recruiting-industry report projects the fastest legal-job growth through 2030 in titles that assume the technical answer won, AI legal specialist, legal technology manager, legal AI trainer, though its methodology was not disclosed and it comes from a site with a stake in listing those roles.
None of this week's items settle which bet is correct. The departure count is small and unverified beyond one vendor's database. The spending figures are commitments, not filled seats. The training camp is one firm's program, not a survey. What is consistent across all four is that nobody is choosing to wait and see.
Exhibit AI
When AI goes wrong in the legal profession. Not to assign blame, but to learn the lesson before it is ours.
A firm called it a strategy change. The gap between that and "AI replaced the work" is where the legal exposure sits.
In late November, Clifford Chance proposed cutting roughly 50 business-services roles in London, about 10 percent of that staff, with around 35 more having their roles changed. The firm's own statement was careful: it was "proposing changes to some of our London-based business professional functions," language that could mean new roles or a reduction, and it did not name AI as a cause.
Reporting at the time understood otherwise. Increased use of the firm's own AI assistant and Microsoft Copilot, alongside expanded offshore operations in India and Poland, were described as contributing factors nobody at the firm would confirm on record. A September 25 legal analysis of the episode argues that gap is not incidental: whether a redundancy counts, legally, as "productive reorganization" or "substitutive automation" changes what an employer has to prove, and most firms making these cuts have not built the internal record to answer that cleanly either way.
The pattern is not new to this firm. A separate Am Law 200 firm's cuts earlier this year drew the same criticism from AI and legal-operations consultants, who argued AI made a convenient explanation for a decision driven mostly by cost and consolidation. The detail worth learning is not which explanation was true. It is that neither firm had the documentation ready to settle the question when asked.
The Lesson
Before a staffing decision goes out the door, write down which category it actually falls into, productive reorganization or claimed AI substitution, and keep the evidence. A vague public reason invites both the legal exposure and the reputational one this reporting keeps finding.