From up or out to up and out

Everyone in law knows the phrase "up or out." It's the profession's defining career bargain: produce enough leverage to justify your compensation and advance, or leave. The pyramid works because the base subsidises the top — lots of juniors doing high-volume, lower-margin work generating the revenue that supports a smaller number of partners doing high-judgment, high-margin work. We described the pyramid economics in Harvey's Strategic Evolution and the training crisis at the base in our previous piece. Here we take a different cut: what happens to the top when the base is automated?

Something is happening to that phrase. A conjunction is shifting. "Up or out" is becoming "up and out." As automation compresses the base of the pyramid, the lawyers who remain are being pushed simultaneously up the value chain, toward higher-order judgment, and out into adjacent domains that were never strictly legal. The two movements aren't separate strategies. They describe a single reorientation of what elite lawyering means when machines handle the technical substrate.

This sounds like strategic evolution. It might be. Or it might be what every disrupted profession tells itself as the ground shifts beneath it.

AI Automates the Base

Julien Bek's distinction between intelligence and judgment, in his Sequoia essay Services: The New Software, captures what's happening. Intelligence work is rule-based, pattern-based, high-volume — contract review, due diligence, first-pass research. Judgment requires experience and taste built over years of practice. AI targets the base because the rules are complex but they are rules. If AI handles the intelligence work, the humans who remain are, by definition, doing judgment work. The question is what that looks like — and whether it's enough.

The Demand Was Always There

Before ChatGPT. Before Claude. The market had already been demanding what "up and out" describes. David has spent a decade documenting it.

The conceptual framework is a shift David describes as the move from "Law's Empire" to "integrated solutions." Dworkin's phrase captured law as an autonomous domain — self-contained, self-referential. That vision shaped legal education, firm structure, and professional identity for generations. But three related crises — a global pandemic, volatile geopolitical conflict, and the eruption of sustainability and social justice issues into corporate life — turbocharged the transformation. Each created problems that were increasingly "legalized" yet extended far beyond what any lawyer could solve alone. Neither the contours of the problem nor the possible array of solutions is strictly legal. Lawyers increasingly work collaboratively with consultants, technologists, strategists, and policy experts on problems that carry legal consequences but require integrated, multi-disciplinary solutions.

The demand for integrated solutions predates AI, and multiple actors tried to build it. Eversheds launched a consulting arm pairing lawyers with management consultants and hired former Big Four partners — though one lawyer asked whether going into consulting was "just dumbing down." But, as David likes to say: "Only lawyers divide the world into lawyers and nonlawyers. And there are a lot more of them than there are of us!" The deeper history of that demand — the Big Four rebuilding their legal ambitions after Sarbanes-Oxley, PwC promoting legal in 124 countries by 2012, in-house counsel like Bridgewater's Tracey Yurko building "virtual law firms" that cut costs by more than 50 percent and concluding that the Big Four held the advantage — is the subject of our competitive-map essay, and we won't re-tell it here. The point that matters for "up and out" is the pattern those efforts share: the demand for integrated solutions had been building for a decade, and every attempt to meet it ran into the same wall — cultural resistance within law firms, ABA Rule 5.4 preventing non-lawyer ownership, the difficulty of physically integrating different professional cultures, and the impossibility of scaling firms that can't take outside investment.

AI doesn't create the demand for integrated solutions. It removes some of the barriers that blocked it. The co-location problem disappears when everyone is in the same digital workspace. The scaling problem changes when AI agents handle the volume work. The cultural integration problem softens when the technology itself is the common platform.

Up and Out: AI as the Mechanism

What "up" means. The integrative judgment we described in our first piece — the partner drawing together financial, regulatory, and reputational dimensions into a single recommendation — is what "up" looks like. AI handles the technical substrate — research, drafting, precedent analysis — freeing the lawyer to spend more time in the intersection where value lives. When intelligence becomes abundant and cheap, judgment becomes the binding constraint.

In our experiential piece, we described the shift already visible in software engineering — from primary actor to director, coach, and editor. Production becomes fast and cheap; the bottleneck shifts to orchestration. That shift from production to orchestration IS "up." If the pattern holds, law is next. It requires all lawyers to become managers who are able to direct, coach, and edit output by AI using their domain expertise, taste, and judgment.

What "out" means. Start narrow: deal structuring that integrates tax, regulatory, and commercial strategy simultaneously. Governance advisory spanning board dynamics, reputational risk, and stakeholder management. Sanctions compliance requiring not just the legal rules but the geopolitical logic behind them. Then expand: the "legalization of everything" from our first piece generating demand at the intersection of law and the full complexity of the client's world. There's also a dimension of "out" that expands law's centrality: as autonomous AI agents proliferate, in-house teams will increasingly govern how organizations use them — liability frameworks, regulatory expectations, risk tolerance. The legalization of AI creates a new and expanding domain of "out."

Harvey Shared Spaces is an example of building infrastructure for this — law firms, consulting firms, and corporate clients in the same AI-powered workspace on the same matter, doing digitally what Eversheds tried to do physically and Bridgewater tried to orchestrate manually. In our reading, Harvey isn't aiming to create "legal AI" per se. It's aiming to create integrated solutions AI — with law as the entry point. The question isn't whether Harvey is a good AI legal tool. It's whether the gateway position we described in our first piece leads to the integrated solutions model that David has been documenting for a decade — and whether AI infrastructure is what finally makes it possible.

The third path gets a name. In our first piece, we described three paths that technology creates for any profession: doing the same thing cheaper and faster, expanding demand through lower costs, or doing fundamentally different things. "Up and out" is the third path made concrete. And the copilot-versus-autopilot binary that Bek frames assumes two categories — the augmented human and the replaced human. But "up and out" suggests a third: the orchestrator. Someone who commissions autopilots, works with copilots, and exercises the integrative judgment that neither can replicate. It's what the best general counsel — people like Yurko — were already doing before AI existed. AI doesn't replace that orchestration function. It makes it faster, more scalable, and more powerful.

The Christensen Question

Clayton Christensen's core insight: incumbents facing disruption from below almost always respond by moving upmarket. They cede the low-end work — it's low-margin anyway — and retreat to the high ground where their expertise and brand still command premium prices. They often believe they're safe there. They're usually wrong. The disruptors improve. They move upmarket. The low-end work was the wedge; the high-end work was always the target. Bek says this explicitly: "The outsourced task is the wedge. The insourced work is the long-term TAM [total addressable market]." Steel minimills started with rebar and ended up making sheet steel. Kodak moved to professional film as consumer photography went digital. Each time, the incumbents pointed to the quality gap. Each time, the gap closed.

"Up and out" is the legal profession's version of moving upmarket.

The Big Four have been building integrated solutions capability for decades — consulting, audit, tax, technology, and global infrastructure at a scale no law firm can match. Law firms going "up and out" aren't racing toward unoccupied territory. They're arriving at a party that's already crowded. In 2018, David's research at the Center on the Legal Profession found that many Global 100 firms were aggressively marketing themselves as providing integrated solutions, in order to meet client demands for advisors who "understand their issues, focus on delivering solutions to their problems, and share risks with them." As one Global 100 partner warned at the time: "We have lost good partners to Big Four. They will move up the value chain — no question — and eat our lunch." By increasingly making themselves look like the Big Four, David warned, large law firms might actually be hastening a world in which the Big Four would no longer have to disguise their ambitions.

To be sure, the Big 4 have yet to fully capitalize on this advantage. A combination of their own internal ambivalence (do lawyers largely hired out of law firms ready to cede the "bet the company" cases they used to pursue to their old law firms while they concentrate on process and technology driven "run the company matters? and the fact that law remains a relative rounding error on the Big 4's balance sheets (does "tax and legal" really want to invest tens of millions to compete with established law firms when tax and advisory are booming?), and the unexpected upturn in the post-covid legal market that produced both surging demand and a flight to quality for high-end services — have largely kept the Big 4 at the margins of the Big Law game, at least in established markets. But as Christensen would undoubtedly note, this temporary surge in demand for incumbents may only mask the underlying disruption building in the market. And the arrival of AI and big data may be just the adrenaline shot these still slumbering giants might need to shake off their ambivalence and status envy and embrace their true competitive advantage to integrate solutions that leverage both process and efficiency at the low end and multidisciplinary expertise at the high end.

And now a fourth competitor for that ground materialised in 2026, one that didn't exist when this convergence began. In a single stretch of April and May, the foundation-model companies stood up dedicated deployment arms with private capital and the big consultancies: Anthropic's $1.5 billion joint venture with Blackstone, Hellman & Friedman, and Goldman Sachs; OpenAI's $4 billion deployment company, backed by TPG with McKinsey, Bain & Company, and Capgemini taking equity, staffed from day one by its acquisition of the applied-AI consultancy Tomoro. All run Palantir's model — forward-deployed engineers who live inside the client's complexity rather than shipping software and leaving. They sell neither tools nor billable hours but deployment: the integrated, cross-domain delivery that "up and out" describes, built from the start on an AI cost structure and carrying none of the partnership overhead. Goldman's Marc Nachmann describes the aim as "democratising access to forward-deployed engineers" — putting integrated AI delivery within reach of companies that could never afford the fees of a Big Three consulting firm. The party law firms are arriving at didn't just get more crowded; one of the new guests is purpose-built for exactly the territory they're trying to reach. And, as the competitive map showed, capital itself is now buying its way directly into the provider's chair — a cap-table software company acquiring a regulated law firm and announcing its intention to move up from routine work into the advice that has always been the incumbents' fortress.

Meanwhile, general counsel aren't waiting — they're building their own orchestration capability, assembling integrated solutions without law firms as the intermediary.

So "up and out" faces convergence from four directions simultaneously. From below: AI-native firms eating technical work, accumulating data, improving. Laterally: Big Four adding legal to their existing integrated infrastructure. From the client side: general counsel building their own orchestration layer. And from the AI substrate itself: foundation-model deployment firms selling integrated solutions at a cost structure no partnership can match.

But Law Might Be Different — or Might Not

We genuinely don't agree on the answer.

David has documented law's structural defenses for decades and believes they're real. The gatekeeper function isn't just a market position — it's regulatory, embedded in how modern society operates. Bar rules, opinion letter requirements, court access restrictions, and licensing regimes create structural moats that technology alone can't breach. A Sullivan & Cromwell opinion letter carries institutional weight no AI platform can replicate. Trust at the level of bet-the-company decisions can't be disintermediated. Regulatory embedding runs deeper than most people appreciate: lawyers helped write the rules that require legal interpretation, and the legalization of everything keeps expanding this. AI regulation, ESG mandates, sanctions regimes — each new domain generates complexity that requires legal navigation.

There is a further point that goes to the pace of all this, not just the direction. Law is not software. A software market reconfigures at the speed the technology allows; a legal market reconfigures at the speed its people permit. The lawyers and the other professionals inside these firms have far more power than their counterparts in most industries to slow, stop, or redirect change — through regulation, through partnership structures, through professional culture, through the simple fact that the people whose judgment the system depends on are also the people who decide how fast it moves. That friction is real, and it is part of why every prediction of the profession's collapse has so far been early. The convergence we have described is arriving more slowly into law than it has into less protected industries, and it will keep arriving more slowly.

And law's structural position generates a dynamic that complicates the Christensen pattern: when AI increases the volume of automated contracts, it also increases edge cases requiring human judgment. More automated compliance at scale means more strategic questions about which risks are tolerable, which trade-offs to accept, how to position across jurisdictions with conflicting requirements. If automation feeds the apex — creating demand for the integrative judgment "up and out" produces — then the relationship between base and apex isn't substitution. It's complementarity. This is genuinely different from Kodak's film. Digital photography didn't create more demand for professional film. But automated legal intelligence might well create more demand for legal judgment. And lawyers as the AI governance layer is an entirely new category of work that didn't exist before AI.

Anthea sees a different pattern. Every disrupted profession thought it had structural defenses — accounting had licensing, medicine had certification, journalism had editorial judgment. None proved as durable as practitioners believed, because disruption routed around them. Friction is not the same as immunity. The structural power that lets lawyers slow change buys time; it does not buy a stop. It slows the reconfiguration without settling it — which is exactly why the competitive map scrambles rather than resolving, and exactly the condition in which an incumbent can mistake a delay for a defence. AI-native legal platforms don't breach the bar; they maintain lawyer oversight while delivering AI-native economics. And the convergence thesis — "today's judgement will become tomorrow's intelligence" — keeps advancing the frontier. Tasks that were considered pure architectural judgment eighteen months ago in software engineering are now handled by AI agents. The boundary keeps moving. The regulatory barriers are already eroding at the edges: Arizona granted KPMG a law license, Utah created a regulatory sandbox, and, as the competitive map traced, the May 2026 wave — Claude for Legal, the deployment joint ventures, a software company buying its way into a regulated firm — kept pushing on the wall while the lawyers debated how high it was.

The honest answer: we don't know. Law's structural position — its regulatory embedding, its gatekeeper function, its century of building itself into the intersection of everything — may make it genuinely different from every profession that believed it was different. Or it may make it the highest-profile example of Christensen's pattern, where the friction that feels like protection is only the lag before the break. What we can say is that "up and out" is not obviously safe just because it feels like strategic evolution, and not obviously doomed just because the upmarket party is crowded. The boundary keeps moving; the question is who is moving it, and how fast the people inside the profession will let it.

The Shape That Emerges

If "up and out" succeeds even partially, what shape replaces the pyramid? Several possibilities.

An obelisk. The same basic shape — hierarchical, vertical, up-or-out logic intact — but dramatically narrower. AI handles the volume work, so the firm takes in vastly fewer people at the base. The rungs are still there; there are just far fewer people on them. It's the pyramid's answer to automation: keep the structure, shrink the intake.

A barbell. A different configuration entirely. Senior partners at the top — the judgment layer, the client relationships, the institutional weight. AI-native juniors at the bottom — fluent in the tools, cheap to deploy with AI leverage, productive from day one in ways previous generations weren't. And in the middle? A hollowed-out, missing generation. The mid-career lawyers who used to spend a decade on progressively complex work, building the integrative expertise that "up and out" demands — they're the ones squeezed out. They don't have the relationships and the judgement to be at the top, and they don't have the AI-native skills to be at the bottom. But over time, this becomes a structural problem because the pipeline of producing people that move from the bottom to the top is broken.

A diamond. Thick in the middle where human-AI collaboration is densest, thin at the automated base and the pure-judgment apex. This is the shape that looks most promising — a wide band of orchestrators, reviewers, and integrators working alongside AI, with training reimagined through orchestration rather than volume work. But it masks a coordination failure. Each firm's decision to automate junior work is individually rational — cheaper, faster, more competitive. No firm needs to train lawyers; it needs trained lawyers. When every firm free-rides on the training pipeline, no one invests in maintaining it. The resource disappears. It's a classic tragedy of the commons applied to professional development. And it gets worse: the verification capacity needed to catch increasingly subtle AI errors is exactly the domain expertise that the collective automation decision is destroying. We're building a system that depends on human judgment we may be setting ourselves up to systematically fail to produce.

A T. This is the shape that emerges if automation doesn't just compress the base but keeps moving upmarket — intelligence work eating into what used to be judgment work, the automated box growing larger and encroaching higher. What remains is a small band of individuals who sprout out on top, extending horizontally: orchestrating the AI-driven work below while connecting outward to clients, other firms, regulators, consultants, and all the adjacent domains that integrated solutions require. The vertical stroke is the automated stack. The horizontal bar is the human layer — thin, stretched wide, doing the relational and integrative work that machines can't.

We don't know which shape wins. We suspect different practice areas and market segments will settle on different structures — and that the answer will differ in New York and London and Singapore and São Paulo.

"Up and out" is the career-level version of integrated solutions. And integrated solutions requires the broadest, most demanding form of the domain expertise that our previous piece argues is being destroyed. It's not enough to be a good lawyer. You need to understand business, strategy, regulation, technology, geopolitics — and hold them in productive tension while exercising judgment about which dimension matters most for this client, this problem. That kind of integrative judgment requires more apprenticeship, not less. More years working on the M&A deal where the employment issue became the deal-breaker, the cross-border transaction where geopolitical context changed the entire risk calculus.

The profession's response to automation — go up, go out, become an integrated solutions provider — depends on a production process that automation is eliminating. The automation that makes "up" possible also destroys the pathway to the breadth that "out" requires. Our previous piece asks: can the profession still produce people with domain expertise? This piece asks: even if it can, can it produce people with integrative expertise — the broader, harder thing?

Whether the profession can go up and out while the ladder behind it collapses is the question this piece leaves with the reader.

We've now mapped the landscape from multiple angles — law's structural centrality, platform strategy, the experiential gap, competitive dynamics, the training crisis, and where the profession goes when the base is automated. In future pieces, we'll take the analysis deeper — mapping the actors reshaping the legal market, the driving forces behind their strategies, the feedback loops that amplify or constrain change, and the scenarios that emerge from different combinations of these forces. The system is in motion. Understanding it requires more lenses, not fewer.