Fourteen legal professionals. Two and a half hours of intensive discussion. Three questions, each building on the last: Which AI-enabled workflows have already proven effective? Which emerging capabilities are poised to reshape the legal industry? And where can young legal professionals find their path to differentiation and breakthrough?
What follows is a distillation of some of the key insights from the discussion, shared as a reference for the legal community.
Three Facts, One Conclusion
Fact One: Lawyers Are Becoming Orchestrators
A lawyer from a boutique firm demonstrated her workflow: AI drafts contracts, analyzes clauses, and conducts compliance checks simultaneously. Her role is increasingly becoming that of an orchestrator—someone who thinks several steps ahead while machines execute multiple tasks in parallel.
“What used to take me five hours to draft now takes ten minutes. So while AI is doing the work, what should I be doing?”
This is not simply a story about efficiency. It raises a fundamental question about how legal organizations are structured. When AI multiplies an individual lawyer’s productive capacity several times over, the traditional resource allocation model of law firms—taking on matters based on headcount and charging clients based on billable hours—begins to shift at its very foundations.
Fact Two: What Truly Sets U.S. Law Firms Apart Is Not Technology, but Governance
The Beijing office of a U.S. law firm has deployed a fully localized AI system that allows its lawyers to draw on the firm’s transaction database spanning three decades to assist with clause drafting. That in itself is hardly surprising—technological advantages can eventually be replicated. What deserves greater attention from Chinese law firms is the governance framework behind the technology: every engagement letter includes provisions authorizing the use of AI, and lawyers must enter the relevant matter number each time they use the system to verify the client’s authorization status. AI adoption without a governance framework does not build competitive advantage—it accumulates systemic risk.
Fact Three: A One-Person IT Department
One independent practitioner, with no technical background whatsoever, used natural-language instructions to have AI build a local data anonymization tool for him—supporting multiple file formats and allowing anonymized content to be restored with a single click. This is not really a story about technology. It is a story about organizational structure. When technological infrastructure that was once a capability reserved for large law firms becomes accessible at the individual level, a fundamental question emerges: What, then, is the true value anchor of the law firm as an organization? Is it the premium attached to its brand? Its ability to aggregate clients and matters? Or, increasingly, is it simply the bargaining power that comes with scale?
One Conclusion: Of the three layers in a law firm’s pyramid—information processing, judgment, and persuasion—the middle layer is being irreversibly compressed.
Information processing—researching laws and regulations, organizing facts, and comparing precedents—is an area where AI has already come close to achieving an overwhelming advantage. Persuasion—winning a client’s trust, bringing a counterparty to the negotiating table, or convincing regulators to accept a proposed solution—remains beyond AI’s reach. Of these two layers, one is clearly losing ground, while the other remains firmly defensible.
What is truly being squeezed is the middle layer: judgment.
This does not mean that “judgment” as a whole is being replaced. AI still cannot make the ultimate business decision. It does not know how much litigation risk a particular client can tolerate, what the unwritten rules of a particular industry are, or when to remain silent at the negotiating table and when to walk away. But the raw material on which judgment depends is being reshaped by AI. When a lawyer with three years of experience can use AI to achieve, within a matter of hours, the breadth and depth of analysis that once required eight years of experience, the scarcity of “judgment” no longer comes from a monopoly on information, but from the value choices made on top of that information.
This violently disrupts the traditional talent development path within law firms—a linear progression from research to analysis to judgment, with capabilities accumulated step by step. One lawyer at the discussion offered a precise summary:
“What the boss is selling isn’t your due diligence work. Anyone can do that. What the boss is selling is judgment and trust. But if you can no longer charge for the execution work and all you have left to sell is judgment—how much can you really charge for judgment?”
This is not a lament. It is a pricing question. The current law firm billing model is, in essence, a form of bundled pricing: the research hours of junior lawyers, the analytical hours of mid-level lawyers, and the judgment of partners are packaged together into a time-based bill. Once AI unbundles the first two layers, that “package” falls apart.
But what is falling apart is only the old structure. Where is the new pricing model? Whoever finds it first may gain the bargaining power for the next decade.
The Client-Side Reality Check: We No Longer Pay for Information Processing
The only client-side voice in the discussion came from the head of legal and compliance at a state-owned enterprise. Her remarks provided perhaps the most valuable reality check of the entire discussion—offering both reassurance and a warning.
Her company is currently moving forward with a localized AI deployment, but progress has been “much slower than expected.” The obstacle is not the technical threshold, but integration—connecting AI with decades of accumulated internal policies, the systems and workflows of different departments, and a complex compliance framework. The destination may look simple, but the road leading to it is full of pitfalls.
Then she said:
“External lawyers remain indispensable for now. But what we need is not for you to spend large amounts of time researching laws and regulations. What we need is your judgment on complex issues.”
“Remain indispensable” is the reassurance; “judgment on complex issues” is the substance. Clients are not deciding whether or not to buy legal services. They are deciding which part of a lawyer’s services they are willing to buy. They no longer intend to pay for information processing—the value of that work is being driven toward zero by AI. But how much they are willing to pay for judgment is something the market has yet to determine.
Three Structural Opportunities for Young Lawyers
Opportunity One: Become a Partner in the Client’s Risk.
AI can tell you that “courts supported a particular strategy in 70% of similar cases,” but it will never tell the client, “You should litigate rather than settle,” or vice versa. That is because “should” is not a legal judgment. It is a business judgment, a judgment about risk appetite, and an understanding of the client’s interests as a whole. During the discussion, a senior partner at a leading “Red Circle” law firm shared an example. One of his clients comes to him whenever a difficult commercial dispute arises, even when the matter falls outside his area of expertise. “I know you think this is not your specialty,” the client told him, “but you are someone who thinks about things from the perspective of my interests. I don’t trust anyone else.” The reason this kind of relationship cannot be replaced by AI is not that it involves exceptionally deep legal knowledge—quite the opposite. It is because, at its core, this is not a knowledge service. It is a form of risk sharing.
Young lawyers should ask themselves: Is my client paying for my legal advice, or entrusting me to share the risk of their decisions? The gap between these two roles is the gap between being replaceable and being irreplaceable.
Opportunity Two: Turn Tacit Industry Knowledge into a Competitive Moat.
The barriers built on pure legal knowledge are being rapidly eroded by AI. But the combined barrier of deep industry expertise + legal knowledge is, in the short term, almost impossible for technology to penetrate. The reason is simple: tacit industry knowledge—how to value a drug approval, how centralized procurement policies affect transaction pricing, or the pricing logic of carbon emission allowances in chemical-sector M&A—is inherently absent from public texts and from the training data of any large language model. It exists only in the minds of people who have spent years immersed in the industry. During the discussion, a lawyer specializing in a niche area of international law shared her own practice. She used AI to organize the non-public regulatory rules, operating practices, and case database in her field into a dedicated search system. “It’s like my field’s own PKULaw.” The tool not only improved efficiency, but also helped two partners find new clients one after another.
The best competitive moat for a young lawyer is to choose an industry and put down roots in it for five years. Let AI handle the legal part for you, while you use your own mind to absorb the industry’s tacit knowledge. Five years from now, you will be the person in that field whom AI cannot replace.
Opportunity Three: Fill the Organizational Vacuum in AI Management.
One awkward fact came up repeatedly: the people in law firms who are best at using AI are often junior lawyers—they are willing to experiment, have the time to experiment, and have no sunk costs. But junior lawyers have no authority to decide whether the team should use a particular tool. Decision-making power rests with mid-level lawyers, who happen to be the group with the least firsthand experience of AI and the lowest willingness to use it. This creates a huge organizational vacuum: Who decides which AI the team should use? Who determines the quality boundaries of AI-generated output? Who establishes compliance standards for AI use? Who designs the workflows and maintains them over time?
In fact, forward-looking mid-level lawyers are actively stepping into this vacuum and positioning themselves as their teams’ “AI architects.” They may not know how to code, but they know which tools are appropriate for which scenarios, how to train and configure AI to match their teams’ document standards, and how to explain to clients that “this document was prepared with AI assistance.” Five years from now, lawyers who serve as these “architects” will be scarcer than today’s salaried partners—not because they are smarter, but because they are willing to step into territory where no one else has ventured.
Distillation
As the discussion drew to a close, a participant with a computer science background brought up a technical concept that could not be more fitting as the conclusion to this article.
“Distillation. The capabilities of a large model can be distilled into a smaller model. When the tacit knowledge of the best lawyers is extracted and encoded by technology, and then distributed to everyone in the form of tools—the chain of knowledge transfer is short-circuited.”
Translated into the everyday reality of law firms, this means that experience that once took five years of learning under a mentor to begin to grasp may, with AI, become accessible to a lawyer with only three years of experience. It is not because AI is smarter than the mentor. It is because the mentor’s experience has been distilled by technology into infrastructure that anyone can access.。
So the discussion ultimately converged on this conclusion: the law firm’s three-layer pyramid will not collapse as a whole, but its middle layer—the analysis layer—is being irreversibly compressed. At the top, the ability to make decisions and exercise judgment, and at the bottom, the ability to master and deploy tools, will become the two poles of the new structure.
Those who can evolve from information processors into partners in their clients’ risks will occupy an irreplaceable position in this new structure. Those who build a competitive moat around the tacit knowledge of a particular industry will find that AI cannot break through their barriers. And those who proactively step into the organizational management vacuum will become critical nodes that cannot be bypassed.
As for those who remain in the middle layer, still doing what AI can already do—
They will not be replaced by machines. They will be replaced by people who started from the same place but have already used machines to accelerate their own growth curves.
The source of anxiety should never be “how will machines change?”
The real source of anxiety should be this: people your age are already using machines to do what you are still doing manually.
