Obsidian/X-Articles/The 10x Lawyer.md

title, source, author, published, created, description, tags

title	source	author	published	created	description	tags
The 10x Lawyer	https://x.com/zackbshapiro/status/2031717962948690355	Unknown	2026-03-11	2026-03-14		clippings

The Great Re-sorting of the Legal Market

Disruption, in the way Silicon Valley uses the word, implies a new entrant eating an incumbent. In the context of the legal profession, this would look like a tech startup replacing a law firm, or a chatbot replacing a lawyer. But that is not what is happening. What is happening is quieter, more structural, and far more consequential: the same lawyers, the same clients, the same work, are re-sorting around a new variable.

For forty years, the legal market has been sorted by institutional prestige. Am Law rankings. Firm letterhead. Profits per equity partner. Clients hired firms, not lawyers, because the firm was the signal. The prestige of the institution served as a proxy for the quality of the individual. That proxy is breaking down.

When production is no longer the bottleneck, the gap between a lawyer with excellent judgment and a lawyer without it becomes visible in ways the old model could hide. The associate who billed 2,000 hours reviewing documents looked productive under the old sort. Under the new one, the question is whether those 2,000 hours produced anything a senior lawyer with AI could not have produced in an afternoon.

The market is about to re-sort around individual practitioner capability. And the re-sort, once it starts, will be fast, demand-driven, and brutal.

The Leverage Machine

To understand what AI is about to break, you need to understand how BigLaw actually makes money:

The economic engine of every large law firm is "leverage:" the ratio of non-equity lawyers to equity partners. In the Am Law 50, that ratio has doubled since 1985, from roughly 1.76 to 3.52.¹

A partner brings in a client. The client's work gets distributed across a team of associates, each billing at rates between $600 and $1,600 per hour, each earning a salary that, while eye-popping by normal standards, represents a fraction of the revenue they generate. The difference between what an associate bills and what an associate costs is the margin that funds partner profits. In 2024, profits per equity partner at Am Law 100 firms averaged $3.15 million, up over 12% year-over-year.² That number is a direct function of leverage.

First-year associates at Cravath now earn $225,000 in base salary.³ By the time you add bonuses, benefits, office space, training, and the lag before they become productive (most firms estimate associates are net-negative for their first year, sometimes two), the all-in cost per junior lawyer is north of $300,000.⁴ But the math still works, because a partner who can keep four or five associates busy billing 2,000 hours each creates a revenue stream that dwarfs the cost.

This is the machine. It has been running, with accelerating intensity, since 1985, when The American Lawyer published the first Am Law 50, ranking the nation's largest firms by profits per equity partner. That single act of transparency detonated the old collegial order. Before 1985, most firms had no idea what their competitors' partners earned. After 1985, they had a scoreboard. Partners could compare their take against their peers at rival firms. Lateral movement exploded. Firms began competing on PPP the way investment banks competed on return on equity, chasing the metric by hiring laterals, increasing associate billing targets, and relentlessly expanding leverage. Average PPP for the Am Law 50 has grown from roughly $300,000 in 1985 to over $3 million today.

Yet the machine contains a vulnerability that nobody had reason to think about until now: the entire economic structure depends on the assumption that legal production requires human labor in rough proportion to the complexity of the work. A complex deal or litigation matter requires more hours. More hours require more associates, more leverage. If that assumption breaks, if complex work can be done with radically fewer hours, the BigLaw pyramid becomes structurally unsound.

The Inversion

In 1968, researchers Sackman, Erikson, and Grant published a study that has echoed through the software industry for half a century.⁵ They measured experienced programmers performing identical tasks and found productivity ratios of 20 to 1 in initial coding time, over 25 to 1 in debugging, and 5 to 1 in program size. The concept of the "10x engineer" was born: the idea that individual capability in software could vary by an order of magnitude, even among professionals with equivalent experience.

The finding has been replicated, debated, refined, and re-confirmed across dozens of subsequent studies over the following decades.⁶ Although the exact ratio is contested, the existence of enormous individual variation is not. And what makes it consequential for software is that the variation is not about effort. It is about judgment: the ability to see the right solution, avoid entire categories of problems, and make decisions that compound in value over the life of a project.

Law never had 10x lawyers. Not because talent doesn't vary (it absolutely does), but because the structure of legal work prevented the best lawyers from delivering returns proportional to their ability. Consider a complex financing with six interrelated agreements and two hundred pages of negotiated text. Three constraints prevented any single lawyer, no matter how brilliant, from delivering outsized returns:

First, cognitive bandwidth: no human can hold two hundred pages of interrelated provisions in active memory simultaneously. The partner with thirty years of pattern recognition still had to read the documents serially, flip between agreements, and build a mental model piece by piece. Their ability to reason across the full deal was limited by how much they could hold in their head at once.

Second, the delegation tax: because the work exceeded any one person's cognitive capacity, it had to be distributed across a team. But delegation dilutes the very thing that makes the best lawyers valuable. The partner's insight about how the one provision interacts with another gets translated into an instruction to a senior associate, who translates it into a task for a mid-level, who marks up the document. By the time the partner's judgment call reaches the page, it has been filtered through two or three humans with less context and weaker pattern recognition. A partner who is 10x better at seeing the issue produces output that is maybe 2x better, because the team compresses the signal.

Third, time as a hard ceiling: a deal that needs two hundred hours of work in two weeks requires a team regardless of individual brilliance. One person cannot do two hundred hours of work in two weeks. The best lawyer's judgment gets spread across whatever fraction of the work they can personally touch, and the rest gets done by people with less of it.

In other words, the variation between lawyers was always enormous, but the production structure of legal work made it structurally impossible to see.

AI removes that ceiling.

A senior lawyer working with AI is no longer constrained by the linear mechanics of production. The precedent review that took a junior associate two days can be done in minutes by the senior lawyer, with roughly the same amount of effort that it would have taken to explain the task to the junior lawyer (perhaps less, if the senior lawyer is using a custom "skill" or "plugin"). The first draft that required a weekend of work product can be generated, reviewed, and refined in hours. The cross-referencing of defined terms across a 200-page deal room, once a task requiring meticulous junior attention, can happen inside a context window that holds the entire document set simultaneously.

This creates, for the first time in the history of the legal profession, the conditions for 10x lawyers. Not lawyers who type faster. Lawyers whose judgment, applied through AI, produces order-of-magnitude differences in output quality and speed. The best lawyer with AI is not incrementally better than the average lawyer with AI. They are categorically better, because AI amplifies the gap between excellent judgment and mediocre judgment in a way that manual production never could.

When that senior lawyer can do with a one-paragraph prompt what a four-person associate team used to do, the leverage math inverts. The associate becomes a cost to the firm that client is subsidizing rather than a necessary income-producing asset.

The Sort

The inversion would not matter as much if clients were unfailingly loyal. If general counsel and founders picked their firms and stuck with them out of habit, institutional relationships would absorb the shock. The BigLaw pyramid would restructure slowly, over a generation, the way industries usually adjust to technological change.

But clients are not unfailingly loyal. They are rational.

There is an old saying in corporate procurement, borrowed from the early days of enterprise computing: nobody ever got fired for buying IBM. In legal, the equivalent has been: nobody ever got fired for hiring Cravath. The prestige of a top firm served a dual function. It was a quality signal to the client, and it was career insurance for the GC who made the selection. If the deal went sideways, the GC could point to the letterhead: we hired the best. That defensive logic has sustained a pricing premium for decades.

A GC evaluating outside counsel has always cared about three things: quality, speed, and cost. Under the old model, all three correlated with firm size. Big firms had big teams, which meant they could staff matters quickly, and their institutional prestige served as a quality signal that justified premium pricing. The IBM logic held because no alternative could deliver comparable quality. Under the new model, a small firm or even a solo practitioner with AI can match or exceed the output of a big-firm team on all three dimensions simultaneously. That has never been possible before.

And when a GC sees it happen on a live matter, the IBM logic cracks. The prestige premium was always a proxy for quality. When a client can observe quality directly, in the work product, in the turnaround time, in the judgment embedded in the markup, the proxy stops being worth what it costs. You do not need the letterhead if you have the receipts.

The re-sort is demand-driven, not supply-driven. It is not happening because firms are voluntarily restructuring. It is happening because clients are moving. When one firm on a competitive pitch delivers the same work product in three days that another delivers in three weeks, the client notices. When the invoice for the first firm is a fraction of the invoice for the second, the client remembers. When this happens on two or three consecutive matters, the client switches.

That switch, multiplied across thousands of client relationships, is the mechanism of the re-sort. Not a technological revolution imposed from outside. A quiet, rational, matter-by-matter reallocation of legal spend toward practitioners who deliver far more for far less. It does not require anyone to believe in AI. It only requires clients to compare invoices.

The evidence is already visible if you know where to look. AI-enabled outside counsel are commanding premium hourly rates, not discounted ones. Corporate legal departments are increasingly structuring outside counsel relationships as fractional, short-term engagements with specialized practitioners rather than ongoing retainers with large firms. The mandate from sophisticated clients is becoming clear: pay for value, not volume. Fewer hours at a premium rate for excellent judgment beats more hours at a standard (or in the case of BigLaw, utra-premium) rate for institutional process.

One Lawyer, One Context Window

Here is what the re-sort looks like at the level of a single matter:

A venture-backed company is raising a Series B. The deal involves a certificate of incorporation amendment, a stock purchase agreement, an investors' rights agreement, a voting agreement, a right of first refusal and co-sale agreement, and a management rights letter. Six documents. Eighty to a hundred pages of negotiated text. Multiple investors with competing preferences on protective provisions, participation rights, and board composition. An aggressive timeline because the lead investor wants to close before quarter-end.

Under the old model, this requires a team of four or five attorneys: a partner for the high-judgment calls (whether to push back on participating preferred, how to navigate the board seat fight), a senior associate to manage document flow, a mid-level to do the first-pass markup, and a junior for diligence and the closing checklist. The client sees a blended rate around $800 per hour and a final bill north of $75,000.

Under the new model, one experienced lawyer loads the entire document set into a context window. The AI holds all six agreements simultaneously. It cross-references the anti-dilution provision in the certificate against the protective provisions in the SPA and the drag-along in the voting agreement. It flags internal inconsistencies that a human reading the documents serially would miss. The lawyer reads the output, applies judgment (this investor historically caves on observer seats but walks away on major investor rights, so push here, concede there), and produces a complete markup with a strategic memo in a single working session.

The client gets better work, faster, for less money. The five-person team was not bad at their jobs. They were simply organized around a constraint that no longer exists: the limitation that no single person could hold the entire transaction in their head at once.

A context window changes that constraint (structurally, not incrementally). The unit of legal collaboration is shifting from conference rooms where humans sync up on what they've each read separately to context windows where a lawyer and an AI co-reason over the entire matter simultaneously. That shift does not require better AI. The AI is already good enough. It requires lawyers who know what to do with it.

The Misdirection

The global legal technology market is projected to reach $65 billion by 2034, growing at over 9% annually.⁷ A significant share of that capital is solving the wrong problem.

The prevailing venture thesis in legal AI goes like this: build the product, remove the human, sell the software. Harvey has raised over a billion dollars across nine funding rounds, at a last reported valuation of $8 billion, with talks underway for another $200 million at $11 billion.⁸ Spellbook, Paxton, CoCounsel, Legora, and dozens of others are pitching some version of the same idea: wrap an LLM in a legal-specific interface, train it on legal data, and sell it to law firms as a productivity tool. The pitch deck always has the same slide. Lawyers spend X hours on tasks that AI can automate.

The pitch is not wrong. It is just missing the point.

The displacement that matters is not happening vertically, from technology to lawyer. It is happening laterally, from lawyer to lawyer. The SaaS product does not win the client. The practitioner who uses AI as a native capability wins the client. Harvey does not take market share from Fenwick & West. A three-person firm that uses the underlying models natively takes market share from Fenwick's five-person deal team, because the smaller firm delivers the same output at a fraction of the cost and a fraction of the time.

The misdirection runs deeper than misallocated capital. It is creating a false sense of security that will accelerate the re-sort for everyone who buys into it. Consider what a legal AI wrapper actually does. It takes a foundational model, Claude or GPT, and puts a legal-branded interface on top of it. It adds a document management layer, a prompt library, maybe some fine-tuning on legal data. Then it charges up to thousands of dollars per seat per month for the privilege. The pitch is that the wrapper makes AI safe and usable for lawyers who do not want to learn the underlying technology.

That pitch is a trap.

A lawyer who uses Harvey or Spellbook as their primary AI tool is, by definition, constrained by what the wrapper allows. They cannot build custom skills that encode their own judgment into reusable workflows. They cannot construct multi-step agentic processes that chain document review into markup into strategic memo. They cannot directly manipulate their work product in Word or Excel through the AI. They are using someone else's idea of how a lawyer should interact with a model, mediated by a product team that, in most cases, has never practiced law. The wrapper becomes a ceiling on what the lawyer can do, and therefore a ceiling on the quality of the output.

The deeper problem is what the wrapper does to judgment itself. When a lawyer's role is reduced to reviewing and approving output that an AI product has generated, the lawyer's judgment atrophies. They stop developing the pattern recognition that comes from wrestling with a problem directly. They become a human signoff machine, a compliance checkbox between the AI and the client. And here is the irony: the moment that signoff function becomes reliable enough to automate (and it will, because it is precisely the kind of rule-based verification that AI handles well), the wrapper company will automate it. The lawyer who outsourced their judgment to a product will discover that they have made themselves into exactly the kind of commodity the re-sort punishes. The company you are paying $1,000 a month per seat is building the system that replaces you.

Conversely, the practitioner who uses the bare metal AI natively experiences the opposite trajectory. Every matter forces them to think about how to structure the prompt, what context to load, how to evaluate the output, where to push back on the model's suggestions and where to accept them. That process is itself a form of judgment development. The AI accelerates the thinking, rather than replacing it. And each completed matter deposits another layer of encoded expertise: custom skills for specific deal types, reusable prompt architectures for recurring analyses, plugins that integrate document manipulation with strategic reasoning. The bare-metal practitioner compounds. The wrapper user coasts.

The models are already good enough. The bottleneck was never the technology. The bottleneck is, and always was, the lawyer.

The Barbell

The re-sort is not egalitarian. It does not raise all boats. It creates a barbell.

On one end: a relatively small number of AI-native practitioners with deep judgment, commanding premium rates. These are lawyers who were already excellent and who adopted AI early enough to compound the advantage. They tend to be mid-career, eight to fifteen years in, with enough reps to have real pattern recognition and enough career ahead of them to justify the investment in learning new workflows. They accumulate work because clients who find them don't leave.

On the other end: a growing pool of lawyers competing on price for commoditized work that AI has made cheaper and faster to perform. Document review. Standard formation work. Template-driven contracts. This work does not disappear. It simply becomes a commodity, priced accordingly. This is the work that the recent batches of Y Combinator "AI-law firms" are going after.

What hollows out is the middle. The BigLaw associates who relied on institutional brand to signal quality rather than developing the judgment that constitutes quality. The mid-tier firms that were neither prestigious enough to command client loyalty nor nimble enough to adopt AI natively. The practitioners who saw AI as a novelty, tried it a few times, got mediocre results (because they wrote mediocre prompts and lacked the judgment to evaluate the output), and concluded it was not ready for legal work.

It was ready. They were not.

The middle also includes the legal tech companies themselves: the wrappers and dashboards and "AI for lawyers" products that positioned themselves as the bridge between the old world and the new one. In the new order, there is no bridge. There are practitioners who use AI natively as part of how they think and work, and there are practitioners who use a SaaS product the way they used to use Westlaw: as a lookup tool, not a collaborator. The first group is producing categorically different work. The second group is producing slightly faster versions of the same work.

The barbell is uncomfortable because it implies that the re-sort rewards people who were already good. That is correct. AI is not a democratizing force in the way its evangelists suggest. It is an amplifier. It amplifies excellent judgment into exceptional output. It amplifies poor judgment into faster mistakes. The gap between the best and the average is widening, and the market is repricing accordingly.

The New Hire

BigLaw's current hiring model is optimized for the leverage machine: hire the largest possible class of associates who meet a minimum threshold of competence and pedigree, bill them hard for several years, and let attrition sort out who stays. The inputs that matter under this model are proxies for minimum competence: law school rank, GPA, law review membership. These credentials tell you that a candidate can read carefully, write clearly, and survive sustained pressure. But they do not tell you whether that candidate has judgment.

Judgment has never mattered all that much to the standard hiring model because it never needed to. The machine runs on volume, not quality. If you hire forty first-years and thirty of them leave within five years, the economics still work, because during those five years each of them billed enough hours to generate significant margin. The attrition was a feature not a bug: extracting leverage from a large cohort, then shedding the ones who would not make partner (or shunting them to "of counsel" or "non-equity partner" roles that preserve the leverage ratio without diluting the equity pool).

After the re-sort, this model breaks. If AI compresses the production work that justified large associate classes, firms will need fewer associates and each one will need to be better. The traits that will matter are the ones the current model ignores: indicia of judgment (clerkships for litigators, non-legal business experience for transactional lawyers), commercial instinct (the ability to understand what a client actually needs, not just what they ask for), social intelligence (navigating a negotiation, reading a room, building trust with a founder who has never raised money before), and native technical fluency (not "completed the firm's AI bootcamp" but "has been using AI tools independently and can evaluate output quality without a checklist").

The irony is that these are roughly the same traits that predict who actually makes partner today. The associates who advance are not the ones with the highest GPAs, or necessarily the ones who bill the most hours. They are the ones with the best business development instincts, the deepest client relationships, and the judgment to handle complex matters without constant supervision. The hiring model just does not select for these traits because it was never designed to. It was designed to fill floors of a pyramid. When the pyramid flattens, the quality of each hire matters far more than the quantity, and the firms that figure this out first will have a meaningful advantage in the re-sort.

The Compounding Loop

The re-sort will not happen once and then stabilize. It will accelerate.

The obvious objection to the re-sort thesis is that experience should favor the incumbents. Cooley and Gunderson and Wilson Sonsini and Fenwick and Orrick have closed more venture financings than any boutique ever will. Their pattern recognition should be deeper, their institutional knowledge richer, their judgment more refined. If reps are what matter, the incumbents should win.

But the objection confuses two different kinds of compounding. At a large firm, experience is distributed across dozens of lawyers, siloed in individual heads, and lost every time an associate leaves. The partner has judgment. The senior associate has some. The junior associate is still learning which end of the SPA to read first. What firms call "institutional knowledge" is really just the sum of individual memories that happen to share a letterhead. The firm's collective experience is vast but fragmented, and none of it is encoded into reusable infrastructure.

An AI-native practitioner compounds differently. Every deal deposits judgment into persistent infrastructure. A custom skill for reviewing NVCA document sets, built and refined across fifty closings, means the fifty-first closing starts from a higher baseline than Cooley's fifty-first closing does, because Cooley's institutional knowledge lives in a partner's head and a junior associate's notes, not in a reusable plugin that can be invoked with a single command. The AI-native practitioner builds agentic workflows that chain contract review into redline generation into strategic memo, each step informed by the encoded lessons of every prior deal. Those workflows are "judgment infrastructure," and they compound with every matter.

This is why the incumbent advantage is smaller than it appears. The question is not who has more total experience. It is whose experience is encoded in a form that accelerates future performance, to the benefit of the client. A firm with ten thousand venture deals in its history but no mechanism to translate that history into reusable AI infrastructure will be outperformed by a firm with five hundred deals and a library of custom skills, prompt architectures, and agentic workflows that make each new deal faster and better than the last.

The firms losing clients experience the inverse loop. Fewer matters means fewer reps for their associates. Fewer reps means slower development of judgment. Slower judgment development means the output quality gap widens. Partners who once kept five associates busy now have three with declining utilization. Overhead stays fixed. The economics tighten. The best associates leave for firms where they can get more reps. The loop tightens further.

This is not a stable equilibrium with winners and losers settling into new positions. It is a flywheel. And it favors the practitioners who started early.

The timeline for the flywheel to become visible is not a decade. It is not even five years. Client-side pressure creates a forcing function that most lawyers underestimate. GCs report to CFOs. CFOs compare legal spend across quarters. When one outside counsel relationship consistently delivers better output at lower cost, the CFO asks why the other relationships are not performing at the same level. That question, asked at enough companies, cascades through the market.

The Shift

A profession that has changed very little in its fundamental structure for a century is about to change a great deal in a very short period of time.

The re-sort will not be announced. There will be no single moment when the old order gives way to the new one. It will happen in the quiet accumulation of client decisions: this matter goes to the smaller firm, this relationship consolidates around fewer counsel, this pitch is lost to someone the partners have never heard of. BigLaw will not collapse. It will thin, slowly and then quickly, as the leverage machine runs on progressively less fuel.

The practitioners who will thrive in the new order are already building. They are not waiting for a product. They are not attending AI seminars hosted by their firms' Orwellian "innovation committees." They are doing the work, matter by matter, building the pattern recognition and the agentic workflows and the client relationships that compound. They are the ones who understood, early, that AI did not change what legal work requires, it just changed how much of it one person can do.

If you are a lawyer reading this, where you end up in the new order need not be determined by your pedigree, your Am Law ranking, or your current title. It is determined by whether you have built the capability to deliver judgment-intensive work at a speed and quality level that the re-sort rewards, or whether you are still relying on an institutional brand to do that work for you.

If you are a BigLaw partner thinking about what your practice looks like in three years, an in-house counsel trying to figure out how to evaluate AI-native outside counsel, a private equity firm thinking about what the law firm of the future looks like and how to build it, or a company at the frontier of AI trying to navigate the legal infrastructure that does not yet exist for what you are building, I am working on these problems every day. I would welcome the conversation.

The sort is underway. The variable has changed. The only question that matters now is which side of the barbell you are building toward.

Notes

¹ Steven J. Harper, "The Belly of the Beast" (blog series analyzing Am Law 50 leverage data). Am Law 50 leverage rose from 1.76 associates-per-partner in 1985 to 3.52 by the mid-2010s. See also NALP, "Law Firm Leverage Drops to Levels Last Seen 10 Years Ago".

² LeanLaw, "Maximizing Profits Per Equity Partner: A Strategic Guide for Mid-Sized Law Firms in 2025" (citing Am Law 100 PPEP of $3.15 million in 2024, a 12.3% year-over-year increase).

³ LawFuel, "How BigLaw Salary Wars and AI Are Upending the Legal World" (December 2025) (reporting Cravath's December 2025 scale pushing first-year base salaries to $225,000, with eighth-years exceeding $430,000).

⁴ BCG Attorney Search, "BigLaw Associate Salaries 2000–2026: Bonus Trends, Regional Variations, and 2026 Forecast" (estimating new associate all-in costs between $250,000 and $300,000 before accounting for overhead and the productivity lag during the first one to two years).

⁵ H. Sackman, W.J. Erikson, and E.E. Grant, "Exploratory Experimental Studies Comparing Online and Offline Programming Performance," Communications of the ACM, Vol. 11, No. 1 (January 1968), pp. 3-11.

⁶ Steve McConnell, "What Does 10x Mean? Measuring Variations in Programmer Productivity" in Making Software (O'Reilly, 2011), ch. 30. See also Fred Brooks, The Mythical Man-Month (1975), ch. 3 ("The Surgical Team"), citing the Sackman findings and proposing team structures built around exceptional individual practitioners.

⁷ Precedence Research, "Legal Technology Market" (2025) (projecting growth from $29.81 billion in 2025 to approximately $65.51 billion by 2034 at a 9.14% CAGR).

⁸ Harvey has raised over $1 billion across nine rounds: a $300 million Series D led by Sequoia (February 2025), a $300 million Series E co-led by Kleiner Perkins and Coatue at a $5 billion valuation (June 2025), and a $160 million round led by Andreessen Horowitz at an $8 billion valuation (December 2025). In February 2026, Harvey entered talks to raise an additional $200 million at an $11 billion valuation. Sacra estimates Harvey hit $195 million in ARR by end of 2025.