by Ian McGullam
ChatGPT landed on our society in late 2022 like an answer-spewing, productivity-boosting, automation-facilitating, some-times-hallucinating bomb. Since then, law firms have raced to harness the productivity gains promised by Generative AI, while eying what it could mean for the broader landscape of legal practice.
We could have dropped the question into a chatbot. Instead, we spoke to four Cornell Law School experts about how the AI Revolu-tion is coming for legal practice—and how much it’s already here.
“The ultimate question for everyone is, ‘Will these technologies replace us? Will they replace the law? That is way too broad of a question,” said Ed Estrada, B.A.’94, the managing partner at Gradient Legal Consulting and an adjunct professor of law at Cornell Tech. “What we know is that lawyers and law firms now have much more advanced tools than they had three years ago, that if they leverage the right way can speed up the work and improve the quality of work they’re providing to clients.”
Generative AI software like OpenAI’s ChatGPT, and other AI chatbots like Google’s Gemini and Anthropic’s Claude, are essentially pattern-recognition machines, using large language models trained on vast datasets to probabilistically generate new answers in response to plain-language queries. Law firms have taken up GenAI tools with a vengeance to automate jobs that once consumed untold hours of junior associates’ time: drafting contracts, doing legal research, ferreting out relevant information from documents turned over during discovery. And it’s showing no sign of stopping. The 2026 Report on the State of the U.S. Legal Market, released in January by the Thomson Reuters Institute and the Georgetown Law Center on Ethics and the Legal Profession, showed law firms investing at record levels over 2025, increasing their spending on technology by an average of 9.7 percent, while spending on knowledge management tools grew by 10.5 percent.
For Daniel R. Alonso, B.A. ‘87, a partner at the New York City office of Vedder Price P.C. focusing on white collar defense and internal investigations, AI streamlines investigative work by helping his team comb through the voluminous sets of Spanish-language documents that crop up often in his work on Latin American cross-border investigations. Alonso himself is fluent in Spanish. But, he says, “By using artificial intelligence, I can handle a matter in a seamless way with non–Spanish-speaking associates. The dataset can be queried in English, and the response can be returned in English.” Alonso, who is also an adjunct professor at the Law School teaching on transnational corruption, went on, “Of course, before we rely on its output, somebody who actually understands Spanish has to look at it. But that’s very similar to cite- and substance-checking, which lawyers are used to.”
Noah Qiao ’15 has made building AI workflows like this his new career, leaving his partner’s office at Kirkland & Ellis in 2025 to found Deepwater & Co. As the company’s AI architect, Qiao helps private-equity companies design AI tools that will both aid business processes and withstand regulatory scrutiny. Qiao, who also teaches classes on private markets, including the relevant AI applications and regulatory considerations, as a Law School adjunct professor, said the stochastic nature of GenAI models—where the same prompts can result in different answers—can result in initially small error risks rapidly compounding. “That’s deadly in enterprise adoption, so managing risks and putting in guardrails is about making the pro-cess robust enough to counteract the risk of hallucination,” Qiao said.
Hallucination isn’t just a concern for Qiao’s private equity clients. In the months following ChatGPT’s debut, news stories began to filter out about lawyers submitting AI-generated motions in court filled with citation of cases and quotations that just . . . don’t exist. AI labo-ratories have sought to tamp down hallucinations like these as they release more powerful versions of their chatbots. However, halluci-nations in courtrooms is still enough of a problem that a French legal researcher maintains a database documenting legal decisions around the world—almost 1,200 as of late March—that have been affected by AI slop. Estrada argues that similar issues are likely affecting transactional legal work—it’s just less exposed to public scrutiny.
In 2024, the American Bar Association, in its first formal opinion on the use of GenAI, allowed its use as a tool but put the responsibilities of competence and confidentiality squarely on the human shoulders of attorneys. And, in the end, the experts we talked to said, keeping those attorneys integrally involved in AI systems is the key to catching hallucinations. “Lawyers are taught on day one how to cite-check, and that’s what we should be doing with generative AI. The lawyers that get in trouble are the ones that forget first principles,” said Alonso.
“Ultimately, if something goes wrong for a client, if there’s a halluci-nation in court, you’re not going to be able to point the finger at ChatGPT, right? You’re a human who is ultimately going to be accountable,” Estrada said. “That is what clients want. They want high quality and accountability. So ultimately, there has to be a human in the mix. There has to be a human verifying, and you have to have confidence in the work product you’re either submitting to a court or delivering to a client.”
All that human verification takes time, however, and that can counterbalance hours saved on more menial tasks. This phenomenon echoes concerns about “workslop,” said Professor Karen Levy, referencing a 2025 study published in the Harvard Business Review finding that companies embracing GenAI were not seeing the productivity gains they expected because employees were creating low-effort output that necessitated extra labor from their coworkers and managers to correct. “It’s not always obvious that it’s the effi-ciency gain that it seems like, especially if that work gets passed up the ladder,” said Levy, an associate professor of information science at Cornell University and an associate member of the Law School faculty who studies how law and technology interact to regulate social life. “Especially in a high-stakes situation like the law, where there are not only professional norms, but actual penalties associated with being wrong.”
Of course, many lawyers who have come to rely on AI tools are confident they’re paying dividends. One Thomson Reuters report said legal professionals surveyed in early 2025 expected AI to save them an average of almost 240 hours a year. But AI’s overall effect on productivity can be frustratingly hard to pin down exactly.
In turn, this complicates debates over the staying power of the billable hour as the legal world’s dominant business model. Hourly rates are continuing a steady trajectory into the stratosphere, with lawyers at the top of the market charging $3,000 or more; the Thomson Reuters–Georgetown report showed double-digit average profitability growth for law firms and a strong demand for legal services amid the economic and regulatory chaos of the second Trump administration’s first year. However, that same report shows increasing budgetary pressure on general counsels trying to afford those legal services.
Are rate collapses or a proliferation of fixed fees coming? Estrada thinks it’s more likely that clients will save by keeping back more work to handle in-house or farm out to the growing sector of alternative legal service providers—with either approach made more doable with AI legal tools. “I think the demand side of the equation is more likely to be impacted before the fee structure is,” he said. “While law firms already offer discounts to clients to secure work or alternative fee agreements, they are unlikely to be the party that offers a wholesale abandonment of the hourly fee structure.”
The AI-powered churn in the legal market has not yet had a clear impact on employment, despite persistent warnings that legal work is highly exposed to GenAI automation. But many in the legal sector express concern over a possible hollowing-out of the apprenticeship logic that traditionally propelled lawyers through the stages of their careers. If AI automates away the lower-level work at a firm, what happens to the associates who would have cut their teeth on that work? “What people frequently say is that a purpose-built LLM can do the work of a summer associate very easily, and that leads us to think, ‘Maybe we don’t need summer associates, or maybe we don’t need a first-year associate,” said Levy. “The issue is that eventually we want those summer associates, or first year associates, to become partners.”
Qiao said the work of early-career lawyers could be substantially different in the AI era, especially as agentic AI—autonomous systems that, unlike chatbots, can act autonomously to achieve complex goals with minimal oversight—becomes more widespread. “The new legal skill sets that a lawyer should really hone is learning to decompose problems and then automate workflows via work-flow logic, vibe code disposable scripts via these agentic models, and then apply a rigorous and zero trust verification standard to every output,” he said. “A partner delegates a task, or you’re on a new deal, and you’re able to apply that process across the board for all the workflows. That’s going to be incredibly powerful, and that’s what’s likely going to power Law Firm 2.0.”
It’s noteworthy that GenAI isn’t the first technological sea change the law has had to wrestle with, though few have moved with such speed. “This is an institution that is built to handle some techno-logical disruption, and we’ve done it before,” said Levy. Electronic discovery, for one, was a key predecessor to the current AI tools radically changing the discovery process, replacing the drudgery of physical paper-shuffling or keywords with algorithms. “Judges and technologists and practicing litigators came together and said, ‘Okay, these are the new norms we’re going to have, and we’re going to reformulate the rules of evidence to reflect responsible use of these tools,” she said. “It’s not exactly the same thing as using generative AI, because it’s a different kind of tool, but it does give me some hope that we can handle this kind of thing if we approach it the right way.” ■
