A paper co-authored by Arvind Narayanan, Sayash Kapoor, and Justin Curl — a third-year at Harvard Law School — argues that advanced AI will not, by default, lower the cost of legal outcomes for consumers, and identifies three structural bottlenecks that stand between AI capability and the transformation some observers predict for the legal industry. The paper was published in Lawfare’s Research Paper Series.
TL;DR: The authors contend that unauthorised practice of law regulations, the relative nature of legal outcomes in an adversarial system, and required human involvement each independently limit how much AI productivity gains can reduce what clients pay.
What it says
The paper applies what the authors call the “AI as Normal Technology” framework — which treats AI adoption as shaped by social and organisational constraints rather than as a function of raw capability — to the legal industry.
The first bottleneck the authors identify is regulation. Unauthorised practice of law rules prohibit non-lawyers from performing legal work, and the paper notes that individuals and organisations face fines and criminal liability if systems are found to cross into the practice of law. Entity-based regulations that restrict who can hold equity in legal services businesses further limit how AI tools can be deployed. The authors write that without reforms, consumers may be unable to access AI legal capabilities regardless of technical progress.
The second bottleneck is the adversarial structure of litigation. Because legal outcomes in the American system depend on relative rather than absolute quality, the authors argue that when both parties become more productive, the competitive equilibrium shifts upward. They write that “achieving the same result — like settling favorably or prevailing at trial — would require a greater quantity and quality of legal work.” They offer digital discovery as a historical analogy: digitisation made document review easier, but litigators used the resulting increase in available digital documents to impose higher costs on opponents, leaving total litigation costs elevated. The authors note that transactional work can exhibit similar dynamics even without explicit adversarial structure.
The third bottleneck is the speed of human decision-makers. The paper argues that if AI enables a flood of legal work, judges will respond by taking longer to resolve disputes or delegating more to assistants, either delaying outcomes or lowering adjudication quality. In transactional work, even AI-drafted contracts require human lawyers to interpret the provisions for their clients, placing an upper bound on how much AI can accelerate legal processes while maintaining human accountability.
The authors also note structural reasons why legal services are expensive: legal services are “credence goods” whose quality is difficult to evaluate even in hindsight, their value is relative, and professional regulations limit competition. They argue these structural factors do not disappear with AI.
The paper offers a conditional conclusion: AI could improve access and efficiency if the legal industry enacts specific reforms addressing each bottleneck, but absent those reforms, the authors write, “we risk a future in which legal work becomes more abundant, but legal outcomes remain expensive and inaccessible.”
The paper is presented through AI Snake Oil, the newsletter by Narayanan and Kapoor, and the official citable version is linked at the Lawfare Research Paper Series.