After Weil, the Pyramid Becomes a Diamond
November 19, 2026 | Intellectual Property News, Legal NewsBy Mark P. Walters
Weil, Gotshal & Manges did not fail because artificial intelligence took its work. It failed the way large law firms have failed for decades.
The firm reported more than $2 billion in revenue for 2025. Its profits per equity partner were $5.593 million. It had no bank debt. On November 10, its partners voted to dissolve it.[1]
Yale professor John Morley studied 37 large law firm collapses for his article, “Why Law Firms Collapse.” He found the same pattern. Partners measure their pay against the firm across the street. A major partner leaves. Revenue falls faster than costs. Every remaining partner then has a stronger reason to leave, and leaving early pays better than waiting. Morley calls it a run on the bank.[2][3]
That is what makes profits per partner a lagging indicator. The number can look strong because it describes the firm before the departures are fully reflected in it. Weil reported $5.593 million in the year it voted to wind down.[4]
The next pressure is different
AI did not cause the Weil collapse. It presents a different problem.
In a 2025 article, Nancy B. Rapoport and Joseph R. Tiano Jr. wrote that generative AI can already perform many tasks at the level of a first-year associate, in a fraction of the time. They concluded that the pyramid model, which uses large numbers of junior associates to perform routine work, may no longer make sense.[5]
A partner departure can be answered. A firm can recruit another partner. A change in the cost of producing the underlying work is harder to answer. Much of the profit in the pyramid came from billing junior time at a markup. When a machine produces the first pass in minutes, that markup is exposed.[6]
The ethics rules already recognize part of the problem. ABA Formal Opinion 512 states that a lawyer who bills by the hour must bill the actual time spent when using generative AI.[7] But a client still cannot see the work behind the bill. The client sees hours charged. The client does not see the first draft, the time actually required, the time written off, or whether an efficiency changed the charge. The meter does not show the saving.[8]
A diamond, not a smaller pyramid
The likely result is not simply a smaller law firm. It is a different shape.
The pyramid becomes a diamond. Fewer lawyers enter at the bottom. The middle holds experienced lawyers who verify the work, make decisions, deal with clients and courts, and exercise judgment. The top narrows to the rainmakers and principal decision makers who bring in the work and take responsibility for it.[9]
Profits per partner may decline if the partnership remains broad. If only a small group is counted as equity partners, the average may remain high. That is why the headline number will not tell the whole story. The important question is how many lawyers share in it.[4]
The pressure will not fall evenly. Rainmakers will continue to be highly paid because their compensation is set by the market for portable client relationships. They should be.[10] The squeeze will fall mainly on the broad middle tier of partners. Their economics depended on leverage below them and on clients financing junior training through hourly bills. Both supports are weakening.
Training becomes the firm’s cost
People age out. A profession that cannot train younger lawyers cannot replace itself.[11]
Under the hourly model, clients financed much of that training without seeing it as a separate item. Junior lawyers learned by doing work that clients paid for by the hour.[12] When machines perform more of that work, the old training system loses both its assignments and its financing.
Firms will have to carry training as their own investment until a younger lawyer is ready for the middle. Some of that investment will be sunk. The lawyer may leave. The firm may never recover the cost. That risk does not make training optional. It makes training a cost of staying in business.
Giving a young lawyer a login is not training. A workable model looks more like an apprenticeship:
1. The lawyer states the issues and her proposed approach before she sees the AI output.
2. She compares her work with the machine’s work and with a senior lawyer’s work, and explains the differences.
3. She verifies citations, records, and factual assertions against primary sources.[13]
4. She attends depositions, hearings, and client meetings, then assesses the decisions made and the results they produced.
5. She takes responsibility for a defined phase of a matter under supervision. Advancement depends on whether she can defend the judgment calls in that phase.
The research supports the concern. In a randomized field experiment involving 133 practicing patent lawyers at 11 intellectual property firms, AI assistance produced larger immediate gains in assisted drafting quality among junior lawyers. On a later unaided task, however, the durable gains were concentrated among senior lawyers. Some juniors used AI as a springboard. Others used it as a cushion. The difference is training.[14]
The buyer will set the pace
Technology alone will not change the model at the speed its promoters predict. Georgetown Law and Thomson Reuters report that 90 percent of legal dollars still move through standard hourly arrangements. The change has to come from the buyer.[15]
It is starting there. A 2025 survey by the Association of Corporate Counsel and Everlaw found that in-house use of generative AI had risen from 23 percent to 52 percent in a year. Fifty-nine percent of respondents said they had seen no noticeable savings from their law firms. Sixty-four percent expected to rely less on outside counsel.[16]
Clients do not need a theory to ask a basic question: if the work took less time, why did the bill not change? Once enough clients insist on a price they can evaluate before the work starts, firms will have to organize around that price.
At Lowe Graham Jones, our answer is phased fixed fees for IP litigation, including patent litigation, in suitable cases and with the approvals our program requires. The price for a phase is set before the work begins. The firm is responsible for how the work is produced. The client is buying the judgment of the lawyers responsible for the matter, not an unverifiable count of the hours spent producing drafts.[15]
The pyramid will not remain the default profit engine of a large law firm. The diamond is harder to build. It requires fewer trainees, better training, and compensation based less on the number of hours billed underneath a partner. It also requires honesty about who bears the cost of producing the next generation of lawyers.
That cost belongs to the firm.
Notes
[1] Reuters, “Weil Gotshal votes to dissolve, ending top law firm’s 95-year run” (Nov. 10, 2026), https://www.reuters.com/world/weil-gotshal-votes-dissolve-ending-top-law-firms-95-year-run-2026-11-10/ ; Reuters, “What’s next for law firm Weil Gotshal as it plans dissolution” (Oct. 29, 2026) (no bank debt, significant cash reserves, and more than $1 billion in receivables and work in progress), https://www.reuters.com/legal/litigation/whats-next-law-firm-weil-gotshal-it-plans-dissolution-2026-10-29/ . Revenue and profits per equity partner figures are The American Lawyer’s, as reported by Reuters.
[2] John Morley, “Why Law Firms Collapse,” 75 The Business Lawyer 1399 (2020), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2580616 . Morley’s study covers 37 large firm collapses since 1988. His core claims: partner ownership “creates the conditions for a spiraling cycle of withdrawals that resembles a run on the bank”; when one partner leaves, the remaining partners bear the loss and become more likely to leave; and partners who stay bear risks, including clawback of distributions, that partners who leave early avoid.
[3] Yale Law School, “Weil Closure Puts Professor John Morley’s ‘Why Law Firms Collapse’ in Spotlight” (Nov. 4, 2026), https://law.yale.edu/yls-today/news/weil-closure-puts-professor-john-morleys-why-law-firms-collapse-spotlight (reporting Morley’s assessment that Weil followed his pattern “almost uncannily according to script”).
[4] “Lagging indicator” is the author’s characterization, not a term used in the scholarship. It rests on arithmetic and on two documented facts. Profits per equity partner is profit for a completed period divided by the number of equity partners, so it describes a past period and rises mechanically when the equity count falls faster than profit. The denominator is a managed variable: equity partners fell from 31.2 percent of firm lawyers (2005-09 average) to 27.8 percent (2020-24), while non-equity partners rose from 14.3 percent to 19.1 percent. Thomson Reuters Institute & Georgetown Law Center on Ethics and the Legal Profession, 2025 Report on the State of the US Legal Market, https://www.thomsonreuters.com/en-us/posts/wp-content/uploads/sites/20/2025/01/State-of-the-US-Legal-Market-Report-2025.pdf . See also William D. Henderson, “An Empirical Study of Single-Tier versus Two-Tier Partnerships in the Am Law 200,” 84 North Carolina Law Review 1691 (2006), https://www.repository.law.indiana.edu/facpub/309/ (testing, and partly complicating, the conventional belief that shrinking the equity tier raises profits per equity partner). Morley’s sourced version of the same point is that collapsed firms remained “formally profitable up through the days they dissolved,” and that relative profitability, not absolute profitability, drives departures. Morley, supra note 2.
[5] Nancy B. Rapoport & Joseph R. Tiano Jr., “Fighting the Hypothetical: Why Law Firms Should Rethink the Billable Hour in the Generative AI Era,” 20 Washington Journal of Law, Technology & Arts 41 (2025), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5080449 (generative AI performs many tasks “at the level of a first-year associate … in fractions of the time”; the pyramid model “may no longer make sense”; firms should shift from billing time and inputs to pricing value and outputs). The article is prescriptive, based in part on interviews with firm leaders; it is cited as argument, not as an employment measurement.
[6] A necessary qualification. The 2026 Georgetown report (cited in full at note 8) found that lawyer headcount at reporting firms grew 2.9 percent in 2025, and it argues that AI may first absorb associate work that was already being written off, an 85.6 percent average associate realization rate, before it touches collected revenue. The 2025 report, by contrast, attributed a reduction in junior associate hiring in part to generative AI. 2025 Report, supra note 4. The claim in this article is about the composition of firms and about pricing pressure as clients demand the benefit of the efficiency. It is not a claim that aggregate associate employment has already fallen.
[7] ABA Standing Committee on Ethics and Professional Responsibility, Formal Opinion 512, “Generative Artificial Intelligence Tools” (July 29, 2024), https://www.americanbar.org/content/dam/aba/administrative/professional_responsibility/ethics-opinions/aba-formal-opinion-512.pdf (under Model Rule 1.5, a lawyer who bills by the hour “must bill for their actual time” when using generative AI).
[8] Thomson Reuters Institute & Georgetown Law Center on Ethics and the Legal Profession, 2026 Report on the State of the US Legal Market: Peak Prosperity and the Fault Lines Below (Jan. 7, 2026), https://blogs.thomsonreuters.com/en-us/wp-content/uploads/sites/20/2026/01/2026-State-of-the-US-Legal-Market.pdf (firms are “deploying technology that can accomplish in minutes what once took hours, then trying to bill for it by the hour”; “clients aren’t eager to see all their productivity benefits flow straight to law firm profits”).
[9] The diamond formulation is the author’s synthesis. For the underlying structural trends, see Marc Galanter & William D. Henderson, “The Elastic Tournament: A Second Transformation of the Big Law Firm,” 60 Stanford Law Review 1867 (2008), https://www.repository.law.indiana.edu/facpub/117/ (growth of non-equity tiers and de-equitization; the equity core increasingly reserved for partners who control access to key clients; shortened time horizons and reduced willingness to invest in training and mentoring); 2025 Report, supra note 4 (shift toward experienced lateral hires and two-tier partnerships). Rapoport & Tiano, supra note 5, likewise discuss firm structures that rely less on large junior classes.
[10] Morley, supra note 2 (relative profitability drives the partner run). On the current market for top producers, see Reuters, “Weil collapse shows fragility behind Big Law’s billions” (Oct. 29, 2026), https://www.reuters.com/legal/litigation/weil-collapse-shows-fragility-behind-big-laws-billions-2026-10-29/ .
[11] Galanter & Henderson, supra note 9 (the elastic tournament reduces firms’ willingness to invest in training and mentoring the next generation). Harvard Law School Center on the Legal Profession’s “AI and the Legal Profession” project (launched Apr. 3, 2026) frames AI as reconfiguring “how lawyers are trained, and what it means to exercise professional judgment,” with a forthcoming essay on “The Training Crisis,” https://clp.law.harvard.edu/research/research-projects/ai-and-the-legal-profession/ . The Harvard project is a research agenda, cited here for framing rather than as a finding.
[12] The classic account of the model in which junior lawyers learn by performing billed work under a promotion tournament is Marc Galanter & Thomas Palay, Tournament of Lawyers: The Transformation of the Big Law Firm (University of Chicago Press 1991). That clients thereby financed training is this article’s economic characterization of that model, not a quotation from it.
[13] Varun Magesh et al., “Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools,” Journal of Empirical Legal Studies (2025), https://doi.org/10.1111/jels.12413 (in a preregistered evaluation, leading AI legal research tools produced hallucinations or otherwise incorrect responses on a substantial share of queries, which is why verification against primary sources remains a lawyer’s task rather than a formality).
[14] David Autor et al., “Does AI Assistance Enhance or Erode Expertise? Evidence from a Three-Month Field Experiment in Patent Drafting,” NBER Working Paper No. 35720 (Sept. 2026, rev. Oct. 2026), https://www.nber.org/papers/w35720 . Pre-registered randomized field experiment: 133 practicing patent lawyers at 11 U.S. intellectual property firms used a custom AI drafting assistant. Assisted drafting quality rose, with larger gains among junior lawyers. On a final unaided task, the advantage was concentrated among senior lawyers; junior lawyers showed no average gain, and their results bifurcated: “AI assistance served as a springboard for some juniors and a cushion for others.” The study was conducted by Google researchers, which readers may wish to weigh.
[15] 2026 Report, supra note 8 (Thomson Reuters Legal Tracker data: “a full 90% of all legal dollars still flow through standard hourly rate arrangements”; the report’s prescribed alternatives include “fixed fees for predictable work, success-based arrangements tied to business outcomes, [and] subscription models for ongoing counsel”). The same report notes the present standoff: more than 80 percent of senior corporate counsel were not yet requiring outside counsel to use AI.
[16] Association of Corporate Counsel & Everlaw, “Generative AI’s Growing Strategic Value for Corporate Law Departments” (Oct. 14, 2025), https://go.everlaw.com/rs/314-QPM-328/images/Generative_AIs_Growing-Strategic-Value-for-Corp-Law-%20Depts.pdf?version=0 (survey of 657 in-house professionals in 30 countries: in-house generative AI use rose from 23 percent in 2024 to 52 percent in 2025; 59 percent reported no noticeable savings yet from their firms; 64 percent expected to rely less on outside counsel). Everlaw is a legal technology vendor and co-sponsored the survey; the figures are reported with that in mind.