Could there be a new Cravath?

One hundred and thirty odd years ago, Paul Cravath organized the modern law firm to meet expanding corporate legal demand. Cravath’s innovative organizational technology eventually diffused to all other major law firms in the US. But Cravath enjoyed flywheel advantages in recruiting and client-facing brand reputation that persisted for decades (there have been minor organizational upheavals since, most notably Kirkland and the rise of the mobile, rainmaking partner).

Today, the legal industry confronts — at slight risk of sounding millenarian — a once-in-a-millennium supply-side shock. Could it present an opportunity for a new type of organization to emerge?

To answer this question, we’ll look at the economics of modern firms and the incentives incumbent firms face. We’ll posit a new entrant, two advantages the new entrant must capitalize on, entry conditions, and how the entrant converts entry advantages to durable organizational advantage, just like Cravath.

Consider an incumbent. Picture Kirkland: huge, full-service, and publicly committed to AI.

Consider an entrant, an AI-native, elite litigation boutique of variable size. We will stipulate only that it’s helmed by an esteemed partner defecting from a top incumbent firm.

Consider a client evaluating the two firms. The client will hire a firm for a matter based on the firm’s expected utility in servicing that matter.

Expected utility depends on (1) which firm produces superior legal work per dollar for that matter and (2) which firm can credibly convince clients of that superiority.

Those considerations in math amount to expected matter value minus the price to service the matter minus the perceived risk of hiring the firm for that matter.

Expected matter value is a function of the value of the particular litigation, or more generally the matter, to the client — on the one hand, bet-the-company litigation, on the other, routine contract dispute — and the quality of the firm’s output applied to the matter.

The perceived risk of hiring the firm is a function of the value of the matter — there is greater perceived risk hiring a new entrant for bet-the-company than for routine contract dispute — and the firm’s brand and relationship with the client.

Quality for a particular matter has four inputs: the firm’s human legal talent, technical production¹, accumulated experience and knowledge, and resources devoted to the matter.

The incumbent holds decisive advantages on all factors but technical production.

It enjoys an advantage in human talent. Risk averse clients select firms relying in part on prestige to signal quality and launder liability, which in turn affects how firms select candidates. Candidates are predictably risk averse and prestige-maximizing in response. They prefer the incumbent, particularly earlier in their careers to maximize optionality and prestige.

It enjoys a decisive advantage in accumulated experience. Accumulated experience depends on per capita human talent and firm size; the incumbent always wins on the latter and initially wins on the former. The advantage is difficult to assail directly by the entrant (though the entrant may attract away important human talent, it will never compete on size).

Finally, it enjoys a decisive advantage in resources devoted to a matter. It wields far more human lawyer-hours and certain types of matters call for significant heterogeneous resources. For example, M&A requires ancillary tax, antitrust, finance, executive-compensation, employment, intellectual-property, regulatory, and sometimes litigation expertise, all of which are more likely to be found within a single incumbent than within a single boutique. The incumbent reduces the coordination cost on behalf of the client, potentially affecting the price per matter term as well.

But the resource advantage is grosser than it is net. Billable-hour incentives reward overstaffing a matter. Origination-credit fights sometimes discourage partners from bringing in other colleagues. The incumbent’s incentives often prevent them from marshaling their absolute resource advantage.

We’ll assume matter types that don’t benefit greatly from the firm bearing coordination costs and that the price term is largely independent from matter value and perceived risk, set at will by the firm.

The incumbent enjoys decisive advantages in both preexisting relationships and brand. However, we’ll assume the entrant is helmed by a well-pedigreed defector from an incumbent who seeds their firm with clients with whom they have preexisting relationships.

The incumbent holds decisive advantages in accumulated experience, resources, and trust, and a weaker advantage in human talent. Clients aren’t moved by a marginal edge in technical production. The entrant needs a decisive advantage.

Why can’t the incumbent just purchase the technical production?

To see why they can’t, let’s decompose technical production into three parts: commodity model capability, proprietary software and data, and adoption/integration capability.

Incumbents and entrants enjoy roughly equal access to commodity model capability. Some incumbent attorneys might not in practice use frontier models because their firms require them to use Harvey or Legora, but nothing prevents them from using the frontier models in the long run.

Firms enjoy modest proprietary software and data advantages. Historically, few firms built their own proprietary software. Some did. One firm famous for patent litigation experimented with in-house-built patent portfolio analysis software, for example. Increasingly, firms commit publicly to increasing investment here (see Kirkland’s $500 million commitment).

Firms accumulate useful data to varying extent, including likelihood of success by phase and cost distribution by matter type and complexity.

Adoption/integration capability favors the entrant. The entrant gains a higher adoption coefficient if it recruits and trains legal personnel with higher complementarity, maintains a smaller personnel count, utilizes talented engineers to boost complementarity, and, more generally, organizes its practice from the ground up around frontier model capability.

Better adoption and integration drives better utilization of frontier model capability and accumulation of proprietary software and data.

There are many frictions that prevent the incumbent from maximizing adoption and integration. These include billable hour cannibalization, organizational inertia, legacy computer systems, risk aversion, and coordination costs across big firms.

The cannibalization concern affords entrants strong counter-positioning. Generative AI, by big-firm attorneys’ admissions, works as well or better than second-year associates on many tasks, but the large firm business model depends on a fat-bottomed pyramid.

Rainmaking partners have no incentive to agree to contract the bottom because they may leave to another firm that won’t require them to, and extracting more value while clients continue to accept the leveraged model does not foreclose later transition.

Incumbents bill an associate hour at a shockingly large fraction of a partner hour despite the associate hour delivering a much smaller fraction of the value. That’s how the pyramidal firm makes a profit. For a long time clients didn’t press on this — they anchored on the headline partner rate, and the leverage of today’s scale simply didn’t exist. Now clients are scrutinizing junior fees directly, and even more so with the sweeping adoption of AI.

Incumbents faced with reducing associate hours would need to quickly coordinate and retrain senior attorneys to effect higher complementarity or increase partner rates and shift business development work to billable work to make up for lost revenue.

These dynamics create a classic disruption scenario beyond mere first-mover temporary advantage.

The entrant must convert frontier model improvements into effective legal technology faster than the incumbent. The entrant does so by better adoption and integration.

Adoption and integration are driven by more effective personnel, less risk aversion (incumbents have been hobbled out of the gate by using lowest-common-denominator offerings — see Harvey’s security positioning), nimble teams, and flexible business models, all set against the disruption dynamics.

Crucially, the entrant must develop proprietary systems to convert frontier intelligence into amplifying legal-technical production across the firm.

Adoption and integration depend in part on in-house engineering talent and systems.

To be sure, the marginal product of engineering labor is proportional to the number of lawyers using whatever the engineer builds, advantaging the incumbent. In the smallest possible boutique, a five-lawyer firm has more difficulty justifying a $500,000 engineer, because that output is only amortized across five lawyers.

But the AI-native entrant’s engineer input may transmit more efficiently, because the entrant designs its systems and talent selection and development specifically to optimize transmission.

Another powerful phenomenon becomes important: complementarity between human legal talent and technical production. In other words, technology makes more experienced or more talented lawyers disproportionately more productive.

The entrant must redevelop hiring, training, and systems around frontier intelligence so it gains a large complementarity advantage.

Marginal gain in output from human lawyer talent as a result of improvements in frontier model capability at the entrant must greatly exceed gains at the incumbent. Likewise, marginal gain in proprietary software and data accumulation as a result of human lawyer feedback must greatly exceed gains at the incumbent.

The entrant’s complementarity engine is proprietary and constitutes a significant part of its competitive advantage.

The incumbent’s strongest advantage is its lower perceived risk, which is a function of the client’s trust in the firm and the value of the matter to the client.

Assume the entrant actually possesses superior matter quality. That isn’t observed directly. Instead, the client observes noisy signals. Traditional branding — Chambers rankings and the like — is unkind to newcomers. Despite a decline in presumptive standing relationships, incumbents still enjoy them to some extent. “Winning” high-value, bespoke work because of a marginal quality advantage is hard to market. To the extent results make up for traditional branding and relationships, low frequency may prevent rapid result accumulation and downmarket results don’t necessarily translate to upmarket matters.

Entrants should pick wedges that best neutralize their disadvantages. They must exploit the traditional branding and standing relationships the entrants’ founding attorneys bring. The client receives more of the high quality partners’ talent, amplified by technology, than they would receive with the incumbent. It would be nearly impossible for an entrant to bootstrap without such founding attorneys.

One possible wedge is high-value portions of large, high-value matters. For example, the client could call in the entrant to lead a series of depositions. The entrant could rapidly process discovery and prepare for depositions using its proprietary technical production. The partner, whom the client would have been delighted to have lead the depositions if the partner were still at the incumbent, should be at least as delighted to have them lead equipped with the entrant’s technical production.

The entrant’s stock of differentiated organizational assets includes:

• Tech: proprietary software, data architecture, evals, engineering capability.
• Talent: conception of AI-complementary legal talent, selection technology, recruiting channels.
• Organization: staffing, training, lawyer-engineer workflows, decision rights, culture.
• Economics: pricing, compensation, R&D allocation, matter selection.
• Brand/distribution: what the firm stands for, public proof, recruiting brand, client acquisition.
• Knowledge: accumulated matter experience and institutional learning.

The entrant’s organizational assets determine to some extent how the entrant exploits the two structural production advantages it enjoys. For example, lawyer-engineer integration, R&D process, knowledge management, and proprietary software increase technology transmission advantage, and a different conception of legal talent, better selection, and differentiated recruiting increase the entrant’s human lawyer talent, which lets the firm exploit complementarity better.

And exploitation accumulates gains in differentiated aspects of the entrant’s organizational assets and capabilities. For example, the entrant selects lawyers with high complementarity potential, maximizes adoption and integration by hiring exceptional engineering talent, takes advantage of its counter-positioned firm structure, and moves faster, unencumbered by incumbent inertia and risk aversion.

The entrant wins a matter if the additional expected matter value the client gains from the entrant over the incumbent, a function of the value of the matter to the client and the firms’ quality measures for the matter, exceeds the entrant’s price disadvantage plus the entrant’s perceived risk disadvantage.

If the entrant exploits complementarity and effectively transmits frontier intelligence, it generates excess rents through increased quality and decreased costs, which in turn generate learning and wins that accrue to various of the entrant’s differentiating organizational assets and capabilities.

As entrant organizational assets and capabilities develop, a new type of organization emerges, defined by:

• Talent: high-information selection, tiny and highly paid associate classes, intensive training.
• Organization: lean staffing and lawyer-engineer teams.
• Economics: proprietary R&D and alternative pricing.
• Brand: public proof of performance.

Two of the most important durable differentiators are the system for identifying and developing complementarity-maximizing human talent and the development of a new kind of brand.

Rents from better exploitation of frontier intelligence must convert into significantly better human talent optimized for the new organization.

The entrant gets better associates. It pays more — for example, $400,000 for first years — because it hires very few. And it hires few because it doesn’t need leverage: technology absorbs much of the work that justified a large number of junior associates. In hiring a few, the entrant can select with far more information than a screener and short interviews or callbacks. The entrant could provide better training because the firm can return to an apprenticeship model — the very model that Cravath displaced — now economical again because machines are doing the menial work. And the technology each associate wields is like having another junior associate at their fingertips. The firm treats associates less as a profit center and more as the future of a good practice.

The pyramid model efficiently organizes expensive cognitive labor. Junior tasks are menial, though important. Selection occurs within the firm, tournament-style. Recruiting is surprisingly low-information compared to other cognitively demanding fields. Initial selection relies almost exclusively on signaling: school, grades, and clerkship. Interviews are primarily vibe checks. The incumbent gets qualified people in the door quickly then aggressively filters them later.

The entrant plays by different rules. Initial selection should consider traditional signals, but also: work samples, simulations, AI interaction, judgment tests, and paid work trials. Elite technology startups have used such techniques successfully for a long time. The entrant’s success depends much more on the character and competence of juniors than does the incumbent’s. The entrant resembles elite technology startups in that regard, whose success depends greatly on their early employee selection.

Incumbent brands are built through prestigious matters, prestigious personnel, client relationships, rankings, lateral/recruiting prestige, and sheer time. Their public marketing is homogeneous: lawyer bios, case announcements, Chambers rankings, client alerts, and conference appearances.

Incumbents built their brand responding to the incentives pyramidal firms faced. They didn’t have differentiated production technology to explain publicly and their promise to clients was reliable throughput of prestige-signaled cognitive effort. A firm with a different production function has something new to communicate, and new methods to communicate it.

Brand for the successful entrant, rather, communicates something about the entrant’s newfound production function and organizational ambition. Development of its recruiting brand outpaces development of its client-facing brand. Juniors flock to the entrant associated with explicit complementarity ambition, unusually selective recruiting, $400k junior compensation pyramidal firms can’t match, tiny teams, rapid responsibility, exceptional training, and a technical, intellectual culture.

Recruiting brand converts into client-facing brand through public proof of the firm’s people and methods: excellent, intellectual personnel who develop their own audiences; empirical legal research, public tools, and benchmarks; demos of methods where confidentiality permits; case studies of wins by elite, tiny teams; and open-sourced non-core technology.

The entrant boutique must resolve not to simply use off-the-shelf, lowest-common-denominator AI products to merely replace expensive, junior associate labor hours. Incumbents already do this to the extent their business model permits (entry-level hiring fell 7.5% for the class of 2025, the second straight annual decline, even as firms grew overall — see Law.com and Above the Law). The boutique may enjoy temporary excess rents compared to the more restrained, pyramidal incumbent. But clients will force incumbents to change more drastically eventually. If the boutique passively uses off-the-shelf, lowest-common-denominator products and fails to engage the organizational flywheel by superior technology transmission and complementarity, it will fail to convert frontier intelligence into durable differentiation.

The Cravath System was organizational technology optimized for elite legal services provisioned by expensive, human cognition. What is the new Cravath System for an economy in which significant portions of cognitive legal tasks are better serviced by machines?

At the turn of the 20th century, law was mostly practiced through an apprenticeship model, in which apprentices would rent desks from a partner, learn from that partner, and bring in their own clients. The apprentice converted the partner’s established reputation into additional clients in exchange for training and use of the reputation. Hiring happened through family connections and networks rather than through merit-based procedures. Partners were mostly solo. There was no systematic training.

Then the rise of Gilded Age businesses required scaled legal services. Paul Cravath capitalized through organizational innovation. The new Cravath would recruit top law school graduates, pay them salaries (itself novel), train them systematically through rotations, place them in an up-or-out tournament, promote partners only from among them, and staff matters with teams rather than specific personalities.

The organizational innovation created recruiting and client-facing brand flywheels. The system was copyable, but diffusion took decades because of reluctance to abandon apprenticeships and nepotistic hiring. Cravath has enjoyed an elite position for over a century, aided — at least in part — by the flywheels it set in motion as first mover.

New boutique entrants benefit from similar flywheels. Their organizational innovations — new firm structure and talent development systems built around frontier intelligence — create very similar recruiting and client-facing brand flywheels Cravath benefited from over a hundred years ago.

Cravath capitalized on demand-side expansion requiring scaled, efficient services, while new boutique entrants today must capitalize on a supply-side collapse in the cost of legal production. But counter-positioning gives new boutique entrants a more powerful head start than the mere inertia Cravath enjoyed. Rainmaking partners at incumbents must voluntarily contract the bottom of their pyramid to compete with the new entrants.

Cravath responded to demand-side explosion by replacing the apprenticeship model with a pyramidal machine that efficiently marshaled the cognitive resources of talented young attorneys. The AI supply-side shock provides an opening for elite boutiques to reimagine firm structure the way Cravath did. Unexpectedly, the reimagining should take the form of a return to a model that looks in size like the one the Cravath machine replaced: a lean firm with hyper-selective recruiting and attorneys maximally augmenting their cognitive resources with AI.