John Nay and the fine print of the future

John Nay once turned down a Fulbright scholarship. He had graduated from the University of Virginia with High Distinction, interned for the Majority Leader of the U.S. House of Representatives, and won an invitation to take the sort of next step that usually requires very little explaining at dinner. He chose graduate school at Vanderbilt instead. The field was computational decision science. The decision would take him from studying how people make rules to building machines that could work with them.

There is a pleasing complication in the career that followed. Nay, now the founder and CEO of Norm Ai, approaches law through engineering. His doctorate came from Vanderbilt’s School of Engineering. His work has nevertheless carried him into law classrooms, legal journals, financial institutions, and conversations about the obligations of autonomous software. The subject keeps changing its clothes. The underlying question has stayed remarkably recognizable: how do you make a complicated human decision intelligible to a computer?

By October 2026, that question has become a business, a research program, and an affiliated law firm. Norm Ai develops agents for legal and compliance work. Norm Law puts attorneys over work performed with those agents. Nay’s ambition reaches beyond helping a professional finish a document: he wants law to help govern AI as it starts doing more things on its own. The fine print, in his account, belongs inside the machinery.

Nearly 70,000 bills, one persistent question

Long before the current enthusiasm for AI agents, Nay was examining the language of Congress. His 2017 paper on predicting lawmaking worked with 68,863 bills introduced from 2001 to 2015. Only 2,513 became law. That imbalance made the assignment more interesting than a simple contest to collect correct answers. A system could predict failure every time and appear impressively accurate, while missing every bill that actually passed. Congress had supplied a rather stern lesson in the limitations of a good-looking score.

Nay’s approach combined legislative text with context, including information about sponsors and their political surroundings. Models trained on earlier Congresses were tested against later ones. The words mattered; so did the circumstances in which those words arrived. Reading the bill and reading the room turned out to be complementary activities. It is an early example of the problem that recurs throughout his work: producing a useful answer requires knowing what the answer is supposed to mean.

He also became CEO of Skopos Labs, co-founded with legal technology professors J.B. Ruhl and Oliver Goodenough. The academic partnerships mattered because the work crossed professional borders. A machine-learning method could find patterns in legal language, but a product needed a reason for people who dealt with law to care. From the beginning, Nay’s projects placed computational techniques alongside people who understood the institutions producing the text.

A classroom, then a regulated business

After Vanderbilt, Nay held a postdoctoral research fellowship at New York University and an affiliation at Harvard. At NYU School of Law, as an adjunct professor, he created the school’s first AI course. He later pursued AI and law research at Stanford CodeX. The itinerary moved between institutions, but the teaching assignment captures something specific about his position: he was bringing a technical subject into a profession whose central working material is language.

There was a commercial education alongside the academic one. Nay was the founding CEO of Brooklyn Investment Group, an investment software platform with an SEC-registered investment adviser subsidiary. That placed AI inside a business where rules were part of daily operations. Building investment technology meant encountering the practical setting in which compliance software would have to earn its place. The distance between an interesting model and a dependable institutional tool can be considerable.

Nuveen announced its acquisition of Brooklyn Investment Group and its parent, Brooklyn Artificial Intelligence, in June 2025. The relationship had begun earlier: Nuveen and Brooklyn entered a strategic partnership in 2023 and developed personalized investment offerings together. For Nay’s biography, the acquisition closes one substantial chapter. By then, he had already begun another. Norm Ai was founded in July 2023, with the legal obligations around AI moving from surrounding condition to central product.

The rulebook gets a working part

The simplest way into Norm’s idea is an ordinary piece of business communication. An investment firm wants to publish material. Someone must check it against the relevant requirements. The task contains numerous smaller questions, exceptions, interpretations, and decisions about what applies. Nay’s approach takes that review apart so that its reasoning can be represented in a form that software can execute and a person can examine.

In his 2025 Vanderbilt lecture, he described decision trees that divide a regulatory problem into constituent parts. Traditional, interpretable systems provide structure; large language models handle judgments that cannot be captured neatly in conventional code. Legal experts build the representation through tools designed for them. A regulatory agent can then evaluate work against the individual provisions. This is a specific design for organizing a review, with expertise built into its steps.

The staffing choice is as revealing as the software. At Norm, attorneys become legal engineers, building and testing agents rather than merely handing instructions to a separate technical team. In an NYSE interview, Nay described how regulations are broken into requirements and nuances before being assembled into a decision process. A lawyer’s interpretation must survive the journey into software. Giving that lawyer a hand in construction shortens the journey.

The intended benefit extends to the explanation. A compliance professional needs to understand why material has been flagged and what requires attention. A verdict without an intelligible route to it leaves the next person with another puzzle. Norm’s design makes the decision process itself part of the product. For an organization answering to clients, internal policies, and regulators, the ability to inspect an answer is a practical feature.

An experiment with consequences

Nay’s research also asks what happens when the same capabilities are used to influence the rules. In a 2023 proof of concept, he tested a language model on corporate lobbying activities. It assessed whether proposed bills were relevant to particular companies and drafted letters seeking changes. The work compared model performance against a baseline and raised a question about how much human oversight firms would retain as the technology improved.

That inquiry complicates the cheerful idea of simply feeding law into AI. A system capable of interpreting legislation might also help shape it. Nay’s paper treated that possibility as a concern for the role law could play in aligning AI with people. If the rulebook is supposed to express human choices, the process by which those choices are made matters. The document and the institution behind it belong in the same conversation.

He was also among the authors of LegalBench, the 2023 collaborative benchmark comprising 162 tasks across six types of legal reasoning. Legal professionals helped construct the tasks. The project asked a narrower, more useful question than whether a model sounds lawyerly: which kinds of legal reasoning can it actually perform? Across these projects, Nay’s public work repeatedly returns to measurement. Fluent sentences need an examination before they receive responsibility.

A law firm joins the experiment

Norm’s expansion adds a twist to the familiar story of selling software to lawyers. Norm Law, the affiliated firm launched in 2025, uses Norm Ai’s technology in delivering legal services. Mike Schmidtberger, formerly chair of Sidley Austin’s executive committee, joined as chairman and partner in January 2026. Attorneys supervise the work. The arrangement places the technology in the same working environment as the professionals accountable for its use.

The economics are part of the design. Norm Law describes pricing around outcomes rather than hours, giving clients a way to benefit from faster completion. Nay has put the organizational question plainly: “lawyers and AI agents need to be fully integrated from the ground up.” It is a proposition about how a firm is built, extending beyond which tool sits on an attorney’s screen.

In July 2026, Norm Ai announced a $120 million Series C at a $1.2 billion valuation, led by Khosla Ventures. Blackstone, Bain Capital Ventures, Coatue, Vanguard, TIAA, and others participated. Those figures describe financing and a private-company valuation. The more revealing connection to Nay’s career is institutional continuity: the buyers and backers operate in the kinds of regulated environments he had already encountered while building investment software.

“lawyers and AI agents need to be fully integrated from the ground up.”

John Nay / On the design of legal services

Keeping the humans in the arrangement

The next research questions grow larger. Norm launched its Legal AGI Lab in April 2026 to investigate the legal infrastructure around AI agents, including accountability, legal reasoning, and how law applies to their actions. The proposed Delaware framework discussed at September’s Central Park AI Forum would test agent-managed companies in a supervised sandbox. It remains a proposal for experimentation, with registration, oversight, and human ownership central to the discussion.

At that forum, Nay appeared with Vinod Khosla and spoke about individual autonomy. His published opening remarks included a smaller, more personal admission: he jokes with his young son and daughter about wanting to freeze them as they are. The parent who wants growth to wait a moment shares a stage with the founder preparing institutions for faster technological change. It is an understandable disagreement to have with oneself.

His stated aspiration is for AI to expand people’s capacity to choose and act while remaining governed by human institutions. That gives the career a coherent arc without pretending its questions are settled. The researcher who once forecast the fate of congressional bills now wants their descendants to guide software acting in the world. For John Nay, a useful future includes someone who can still ask for an explanation - and someone obliged to provide it.