← Latest papers
💰 quantitative finance

AI, Specialization, and the Organization of Knowledge Work: Evidence from Legal Services

This paper argues that while AI lowers the cost of specialized knowledge, its impact on professional service firms is determined by how clients reallocate work and how suppliers respond, leading to a scenario where premium firms can raise markups despite losing market share to lower-priced alternatives and self-provision, as evidenced by trends in U.S. legal services where AI limitations in complex tasks and shifting demand patterns have not resulted in widespread internalization or declining prices for high-end legal expertise.

Original author: Liu Zhang, Mark Esposito, Isabel Yang

Published 2026-11-19
📖 7 min read🧠 Deep dive

Original author: Liu Zhang, Mark Esposito, Isabel Yang

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the author. For technical accuracy, refer to the original paper. Read full disclaimer

When a person faces a complex problem, they must decide whether to solve it themselves, hire an expert, or simply let it go. This choice is at the heart of how modern economies organize work. For centuries, the rule of thumb has been that if you can make something cheaper, people will do it themselves. If a machine can build a chair for less money than a carpenter, the carpenter's shop shrinks. But when the product is knowledge—like a legal argument, a medical diagnosis, or a financial strategy—the rules are less clear. Experts have long argued that artificial intelligence, by making specialized knowledge cheaper and easier to access, would force companies to bring more work inside their own walls, replacing outside lawyers and consultants with their own staff. The question is whether this shift is actually happening, or if the reality is more complicated.

A new study examining the legal system in the United States suggests that the story is not a simple swap of outside experts for inside staff. Instead, the arrival of artificial intelligence has triggered a three-way split in how legal work is handled. Some people are indeed doing more of their own legal work, but they are not hiring fewer lawyers overall; rather, they are changing which lawyers they hire. The research, which analyzed court records, employment data, and the financial reports of law firms, finds that while artificial intelligence has made it easier for individuals to represent themselves in court, it has not caused a mass exodus of lawyers from law firms to corporate offices. Instead, the market has shifted in a way that allows the most expensive, prestigious law firms to charge even higher fees for the complex work they keep, while smaller, cheaper firms see their demand grow.

The researchers began by looking at the most direct effect of artificial intelligence on individuals: the ability to write legal documents without a lawyer. In the United States federal court system, there are two main types of cases where people often appear without a lawyer: prisoner petitions and civil cases involving ordinary citizens. For decades, the rate at which ordinary citizens represented themselves remained steady, hovering between 9.5% and 12.1% of all cases. Then, in late 2022, a new type of artificial intelligence became widely available to the public. By the end of 2025, the share of civil cases with a self-represented party had jumped to 16.8%. This was not a small fluctuation; it was a sharp, sustained rise. The increase was most dramatic in cases that required a person to write a detailed complaint from scratch, such as civil rights lawsuits or disputes over property. In contrast, cases where the paperwork was simple or where a lawyer was already provided for free, such as social security appeals, saw no such rise. This pattern suggests that the new technology is helping people overcome the specific hurdle of drafting complex documents, encouraging them to file cases they might have otherwise abandoned.

However, the study found that this rise in self-representation did not lead to a collapse in the demand for paid legal services. If people were simply replacing lawyers with software, the total amount of money spent on legal help should have dropped. Instead, real household spending on legal services remained above its pre-pandemic trend. The researchers then investigated whether companies were using this new technology to hire fewer outside lawyers and instead build large teams of in-house legal staff. They tracked the number of lawyers working inside companies versus those working in law firms. The data showed no sudden shift after 2022. The proportion of lawyers working inside companies had been rising slowly for years, but the rate of that growth did not accelerate when artificial intelligence became popular. In fact, the share of lawyers working in-house in 2024 was almost exactly the same as it was in 2019.

Why would companies not bring this work inside if the tools to do it are cheaper? The answer lies in the limitations of the technology itself. While artificial intelligence has become very good at answering specific questions or summarizing documents, it still struggles with long, complex tasks that require careful judgment and the ability to handle many steps at once. The researchers looked at benchmarks testing how well these systems could handle real-world legal work. They found that even the most advanced models failed to complete a significant portion of complex, multi-step tasks reliably. Because these tools are not yet trustworthy enough to handle the most critical parts of a legal case without human oversight, companies have not felt confident enough to replace their outside experts with internal teams.

The most surprising finding of the study concerns the law firms that do remain outside the companies. One might expect that if clients are doing more work themselves and turning to cheaper alternatives, the big, expensive law firms would have to lower their prices to compete. The data shows the opposite. In 2025, the largest law firms in the United States reported a 13% increase in revenue, even though the total number of hours they billed remained nearly flat. This means they were charging significantly more per hour. At the same time, smaller, lower-priced firms saw a stronger growth in the demand for their services. The researchers explain this through a process of sorting. As artificial intelligence makes it easier for clients to handle routine, repetitive tasks, those tasks move away from the big firms. What remains for the big firms is the most difficult, high-stakes work that requires deep expertise and a reputation for reliability. Because this remaining work is less sensitive to price—clients cannot afford to make a mistake on a major lawsuit—the big firms can raise their fees.

The study also noted that the law firms most eager to adopt these new tools were often the ones with lower revenue per lawyer, suggesting that smaller firms are using the technology to compete more effectively on the tasks that have been left behind. Meanwhile, the largest firms, which handle the most complex and high-value matters, have been slower to integrate these tools, likely because their clients are more cautious about using artificial intelligence on sensitive, high-risk cases. This caution limits the immediate benefit the big firms get from the technology, but it also protects their position in the market.

Ultimately, the research paints a picture of a legal market that is not shrinking or collapsing under the weight of artificial intelligence, but rather reshaping itself. The technology has lowered the barrier to entry for individuals, leading to more people representing themselves in court. It has not, however, caused a wholesale move of legal work inside corporate walls. Instead, it has pushed routine work toward cheaper providers and left the most complex, high-value work with the premium firms, who have responded by raising their prices. The result is a more divided market where the cost of expertise does not simply fall for everyone, but instead shifts depending on the complexity of the task and the willingness of the client to trust a machine. The findings suggest that while artificial intelligence changes how work is allocated, it does not necessarily make the most specialized knowledge cheaper for those who need it most.

Drowning in papers in your field?

Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.

Try Digest →