As AI moves from experimentation to operations, new models of legal service delivery are beginning to emerge. In this episode of AI and the Future of Law, Jen Leonard and Bridget McCormack are joined by John Nay, founder and CEO of Norm AI and Norm Law, for a conversation about agentic law, AI-native legal services, and what it means to build legal work around agents rather than copilots.
John explains how agentic law embeds legal and regulatory reasoning into AI agents, allowing them to perform first-pass work, learn from prior matters, and support legal teams in ways that compound over time. The conversation explores how this differs from traditional copilot use, how Norm Law combines attorneys, legal engineers, and AI engineers, and why client-specific knowledge can be built into persistent “digital associates” that do not leave the firm.
The episode also examines outcome-based pricing, client trust, the role of experienced legal talent in AI-native firms, and the relationship between Norm AI and Norm Law. John also discusses the Legal AGI Lab, including research into how AI agents may be evaluated, benchmarked, and governed by law.
Key Takeaways
● Agentic law goes beyond copilots: AI agents can perform first-pass work in the background, learn from matters, and improve over time.
● AI-native firms require new roles: Norm’s model brings together practicing lawyers, legal engineers, software engineers, and AI engineers.
● Client knowledge can become persistent: AI agents trained on client preferences and matter history can remain with the client relationship and deepen over time.
● The business model changes with AI: Norm Law focuses on outcome-based pricing rather than billable human hours.
● Human supervision remains central: Norm Law attorneys supervise the work, and nothing goes out without final review and signoff by a barred attorney.
Final Thoughts
This conversation offers a look inside one of the most ambitious experiments in AI-native legal services. John Nay argues that the future of legal AI is not just about faster drafting or better prompts, but about building legal agents, legal teams, and legal business models designed around compounding knowledge, outcome-based value, and the governance of increasingly autonomous systems.
Transcript
Intro + AI Aha!
Jen Leonard: Hi everyone, and welcome back to the AI and the Future of Law podcast, the show where we explore all of the shapeshifting dynamics related to artificial intelligence and how they impact the world of law.
I am Jen Leonard, your co-host and founder of Creative Lawyers, joined as always by Bridget McCormack, president and CEO of the American Arbitration Association.
Hi, Bridget. Lovely to see you.
Bridget McCormack: Hi, Jen. Great to see you too. I’m excited for today’s conversation.
Jen Leonard: Yeah. Today we are joined by a guest for whom we have many questions, because he is at the helm of a really interesting company.
Today’s guest is John Nay, and John is the founder and CEO of Norm AI and Norm Law, the world’s first AI-native full-service law firm. He brings more than a decade of research experience at the intersection of AI and law, with academic appointments at Vanderbilt, NYU, Harvard, and Stanford CodeX. And he created the first AI course at the NYU School of Law.
John, welcome to the podcast.
John Nay: Thank you for having me.
Jen Leonard: Yeah, of course.
So, as you may know, on every episode we ask our guests to share an AI Aha! — something they’re using AI for in their life somewhere that they think is particularly interesting.
We’d love to hear what you’re using AI for these days, aside from the obvious.
John Nay: Yeah, using it for a lot of things.
The one that’s probably the most interesting is around research. As you mentioned, I used to do research. I don’t really have a lot of time for that anymore, but it does help with that to the extent that I can do that.
We launched something recently called the Legal AGI Lab, and the only way that I can have time to get involved in that is to use AI to help with developing research hypotheses and thinking through ways to carry forward the work that we do there.
So it’s sort of like a meta AI legal AI, because it’s using AI to help research legal AI.
Jen Leonard: Love that.
Bridget, I know you do this kind of work all the time too, as a fellow researcher who doesn’t have the time for it.
Bridget McCormack: I do. I feel like it’s the only way I can have my hands in research, even from 10,000 feet, which is the best I can do — to at least hopefully frame the questions that I then want my team to dig deeper into.
But it has turned into such an excellent thought partner for me, maybe in part because it’s available at all of my hours and so patient with me when I say, “No, I didn’t say that right,” or “I don’t think you understood me right.”
I could never do that with a coworker. It would be unfair.
But it is a great way to be able to have my hands in at least a little bit of research on the things that I’m trying to figure out. So I love that.
Agentic Law and the Shift from Copilots to Agents
Bridget McCormack: Well, we’re excited to hear a lot more about how you’re using AI at work. John, your company is a pretty exciting one, at least for Jen and for me, because we talk about AI and law.
But I want to start from the beginning and have you tell us what agentic law means. What does that mean to you? What does it mean to your firm, to your company? And why does it matter that we’re shifting from AI experimentation to AI operations?
Give us the framing of your business.
John Nay: Sounds good.
So agentic law is embedding law into AI agents, to put it really simply. And why that matters, I’d say there are two sides of the coin.
One side is: how can we bring automation to bear for legal and regulatory operations, mainly doing the first pass of some of that work? And I’ll talk about why that matters in a second.
Then the other side of the coin is around governing AI.
Traditionally, the idea of guardrails around AI and governing AI has been relatively deterministic, and it has worked well enough when the systems themselves weren’t agents.
But as the key systems are agentic — so think generating marketing, or talking directly to a client, or giving investment advice as an agent to a client — a lot of stuff that traditionally was human, that core economic activity, needs other AI agents to check that from a legal, regulatory, and compliance perspective.
And if you have agentic law, if you have law embedded into AI agents, that makes that possible. It unlocks the ability to check other AI agents.
So for us, building agentic law unlocks both sides of the coin: automating some of the first-pass work that happens in a more prosaic way, but then also unlocking the ability of AI to be deployed as agents.
Jen Leonard: Interesting.
I have a follow-up question, John, about the agentic approach. What I heard from what you said is making me think of some of the recent news around law firms moving from having their attorneys prompt different LLMs to actually trying to create their own proprietary models with their attorneys as the knowledge layer.
Is that different from the way you think about building agentic legal services? Or is it similar?
John Nay: Yeah. So in that first side of the coin that I was talking about — around agentic law enabling the automation of some of the work that happens inside law firms or inside corporate legal departments — I think there’s a distinction that we’re seeing between the use of AI as a copilot and the use of AI as an agent that can autonomously, in the background and in parallel, be doing processes that are relevant to the work.
When you use AI as a copilot, it’s definitely an efficiency booster. It’s definitely helpful for getting the work done more efficiently.
But after you do the work, it’s sort of like throwaway work. The work that was done is usually very high quality, and it’s great, and the client was happy. It’s all good.
But then the next time you do that work, it’s not like you got that big of a boost from being able to embed that previous work into something that had compounding value — doing it faster and better and cheaper the next time.
Whereas if it’s an AI agent, then you’re more likely to get that compounding benefit because the AI agent is doing more of the full pass of something. You’re able to take the learnings as it’s operating and embed that back in more naturally.
And even before you do the work, the other key distinction is around understanding what it can and can’t do.
When you’re using it as a copilot, it’s less clear because it’s sort of fully merged into the human what the AI can and can’t do. And that makes it harder to trust the AI and to offload something cognitively for the AI to do without having to just recheck everything it’s doing.
Whereas if you have an AI agent, then you can understand the jagged edge of capabilities — what it can and can’t do. And you can benchmark that, and you can systematically study that offline and online.
What that allows you to do is push out the boundary of what it can do by deliberate effort in improving it, giving it more context around the things that it doesn’t know enough about.
Bridget McCormack: Can you make it concrete?
I think most lawyers are now familiar with AI copilots. In fact, most maybe even use them at this point. I don’t know if that’s true or false. Maybe I talk to a different group than others.
But I think they know how to use them to help with research and drafting. They probably have access to a few different copilots at work and probably also at home on their personal devices.
But give a concrete example of what a lawyer using a copilot is doing, and for what kind of work, and then what an agent that your team has built would do — and how that differs concretely.
John Nay: Let’s take the example of doing a deal for a client.
If you’re doing a deal and you’re using a copilot, the docs come in and you’re going to check the docs against maybe a form for that type of deal that you already have. Maybe you have playbooks for that client. You have a term sheet associated with the deal, et cetera.
So you’re checking the docs with respect to those things and looking for issues that you’d want to flag to the client.
If you’re doing that on a copilot basis, then you’re able to put that in and prompt it and see what happens.
If you’re doing that in an AI agent, an agentic way of doing it, that work was actually first just done by an agent.
So things came in the door. Those documents were analyzed by an agent behind the scenes before a human even stepped in at all.
This unlocks a couple of things.
One, it unlocks the ability for that to just happen in the background, in parallel, at any time. The docs could have come in at midnight, and that was running from midnight to 12:35 a.m. It was just happening.
So one thing is just being able to move faster and not have to worry about the bottlenecks of that first pass from a human perspective.
But then number two, we found that it’s able to look across more things and do more things because we’ve pre-trained, in a way, that AI agent to understand the nuances of that particular type of deal, and even that particular client and their preferences and proclivities, and the things that they care about — terms that they would want to push on, things they don’t care as much about.
That’s already baked into that agent’s understanding of that client and that specific type of matter.
Compared to a copilot, where you’d have to re-input that information in a way — you’d have to sort of implicitly train it again about that situation — that’s already been done, and that’s stable.
And it’s compounding, even, because as you do the next deal for that client, it had that previous deal for that agent for that client.
So it’s both faster, and then you can parallel process it. And it’s able to go further up the level of automation because of all that you’ve imbued into it ahead of time.
That makes a big difference in terms of a practical deployment of an agent versus a copilot for a high-stakes matter that you want to move fast on.
Legal Engineers, Client Knowledge, and the Business Model
Jen Leonard: I have a follow-up, John.
You’re obviously a fairly new entrant into the legal services market. And I would think that law firms — one of their advantages, I think — is their document management system, which has decades of their work product and the relationships the attorneys have.
And you mentioned the client preferences that you all know in advance, that you train the agents on so that they can automate the intake.
Do you have engineers who are working to train the models? And how are they integrating client preferences when they’re a blank slate, to some extent, on the work that you’re doing?
John Nay: There are a couple different types of people involved here.
There are legal engineers. This is a career that we developed a couple years ago, where we bring in attorneys — most of them are around the fifth-year associate level from other law firms — and almost none of them have written a lot of code before they join.
Then they come here, and they basically become software engineers.
So their day-to-day is building and refining and testing and deploying AI agents. That’s their full-time job. They’re not practicing law. They’re not doing anything else.
So that’s a key stakeholder in this process.
The second category is software engineers and AI engineers. They can build things that require more traditional software engineering background to stand up: new production-grade harnesses around agents, infrastructure behind the scenes, et cetera.
And then the third category is at Norm Law: the attorneys who are practicing law at Norm Law. That ranges anywhere from senior associates up to very senior partners with decades of experience at other top law firms.
All three are collaborating — especially the legal engineer and the partner — every single day to embed that knowledge that the partners have from decades of experience into these AI agents.
And those agents have a great starting point.
So we haven’t yet talked about the client. We’re just talking about the starting point for that type of agent that represents that matter type, to go deep on that.
Then we come to the client, and we can embed, on top of that base AI agent that could apply to anybody, a client overlay of their precedent and their past deals or past contracts — whatever is the focus for that matter type.
So then it starts to understand and learn about that client.
And going forward, we can learn more and more about the client. The agent overlay on the client level can get thicker and thicker as we move forward and do more and more work for the client.
So that is how we start off. Even though it’s a new situation we have here, we can get pretty far along with those three different components.
Bridget McCormack: And then does your base agent that you train with the client’s specific preferences, or specific prior deal information or contract information — does that client agent then stick with that client?
Do you have a team of a human Norm Law lawyer and an agent that has been specifically trained for that client?
How do you staff your matters when you have all these good choices?
John Nay: Right. So to answer your question, yes.
And this is a great thing, because typically, when you have an associate that has been implicitly trained for that client — to know about them — and then they leave, or something happens to them, then that’s gone.
Whereas here, if you have an agent that everyone has done the hard work to train — the client and the attorney have done the hard work to train that, whether it’s a human associate or a digital associate — now this digital associate is there, and it’s not going to leave.
And so that actually helps a lot for the long-term relationship and making it better for the client.
Jen Leonard: John, Bridget and I do a lot of presentations to law firms and corporate legal departments on what’s happening in the AI landscape, and we have a section on AI-native firms like yours.
The biggest question that we get from law firms or other audiences is: what does the business model look like?
Obviously, you are counter-positioning yourself, I think, to law firms and the billable hour model and the capture of the individual client relationship.
But what is the business model?
John Nay: We don’t price by billable human hours.
We primarily price by outcome. And by looking at whether it’s a deal or an agreement to be reviewed — whatever is the unit of analysis — we align with the client, often based on their historical precedent of what they would pay for something. We align on a price ahead of time.
That way, on a going-forward basis, they don’t need to worry about us devoting a lot of time to it and being very available, because they’re not paying anything extra for that.
And then on our side, obviously, we are incentivized to really lean in as much as possible to improving the service for them and making it faster by using technology in addition to humans.
Bridget McCormack: Has that been controversial with clients?
One of the interesting things we’ve heard sometimes from the client side is that they hate the billable hour and they don’t know how to quit it.
Have there been interesting conversations around how we price this particular matter or this service that have been more complicated than you thought?
How has that gone?
John Nay: I definitely always expected this to be a very interesting problem to solve, because it’s hard a priori to say, “Okay, this type of thing is going to be exactly this much work.”
And there are a lot of times when you want the flexibility of being able to have things meander into more or less complicated situations, and then just be there and be compensated for all that work that you’re doing for the client.
So this is a fundamentally really hard thing to solve.
But it’s one of these where AI just came onto the scene and forced everyone to reckon with trying to solve it, because it just doesn’t align as much with the actual business.
So to answer your question, it does limit the focus, at least initially, of the types of matters that we’re going to do.
We’re not going to be doing things where it’s very hard for both sides to understand ahead of time exactly what’s going to be entailed. It takes that, for now at least, off the table of what we’re going to be doing with clients.
But then it’s also why we build big, long-term relationships with larger clients, because this allows us to really lean in and partner with them on, in some ways, co-developing the way to make sure that we are delivering the value and that the value is being priced correctly.
So it tilts us toward getting these long-term relationships with clients where we’re going to commit to doing a lot together.
And then that allows us to also average out between some things that maybe take longer or take less long. Because if we’re averaging out over more, then both sides have more clarity.
Bridget McCormack: Fascinating.
I’m also interested in how clients have responded. I would love to know more about who your first clients are. You probably can’t talk about that specifically.
But how are clients feeling about trusting the agents you’ve trained to do work that your humans are not in the loop for? Because you’re solving the big problem, right? Having to have a human in the loop at the front end — and often both the front end and the back end — really does limit where AI can go within the business of law.
But you’re solving that with these agents and their harnesses. Are clients comfortable? Lawyers are not comfortable with that for the most part. I guess yours are, but most lawyers are not.
John Nay: On the front end, as you put it, there’s the building out and the testing and the validating. On the back end and on the top, there’s the supervision of the agents and all the work that comes out the door.
So Norm AI is building these AI agents. We’re doing it in collaboration with many lawyers. And then those AI agents are powering a lot of the work behind the scenes at Norm Law.
Norm Law attorneys are supervising that work, and nothing goes out the door without that human supervision by a barred attorney who is doing that work for the client.
So it has that final layer, that final supervision, and final signoff fully baked in on the top.
To your question on the client comfort level, that’s key to it.
And I can say a few of our clients. We have client testimonials actually on our website from a couple of them already.
Blackstone is a client of ours, and so is Coatue. We’re representing large private equity firms like Blackstone and leading hedge funds and investment firms like Coatue.
They’ve gotten comfortable with the team that we have that is able to supervise the work and communicate it to them.
Talent, Client Trust, and the Norm Law Model
Jen Leonard: I have a question, John, about talent.
Building on Bridget’s last question around client trust, it seems to me, based on what you’ve described, that the legal talent you have in-house needs to be exceptionally experienced and have great judgment — all the things that clients are looking for.
What is the challenge like in recruiting top legal talent when lawyers are not always comfortable with something totally entrepreneurial and unknown? Does it make it more complicated?
John Nay: I would say it makes it more complicated, but it’s very doable.
You just have to be clear about what the role is and allow people to self-select into it, because it’s definitely not for everybody. It’s novel and it’s different.
But for those who want to do this — for those who see how AI can have such a positive impact for their clients but are many times blocked at their current firms from being able to pass through some of the benefits of AI to their clients, given the structural constraints of the business model and the lack of innovation at these other places — they’re like, “Oh, this is an amazing outlet for me to be able to express that interest and align it with a business model.”
So we are finding that a lot of people are very excited about that.
And then the unique ability to collaborate with AI engineers and legal engineers every day, and be part of this project of building lasting value and building AI agents that understand law, is just a really fun project to work on.
It’s very intellectually satisfying, and you get the gratification of building something lasting and compounding. So that’s been a huge draw for a lot of people as well.
A few people are worth calling out. Mike Schmidtberger is the chairman of Norm Law. He ran Sidley Austin for the past seven or so years before doing this, and now he’s full-time partner and head of investment funds and regulatory at Norm Law, as well as chairman.
We’re getting people like that. We also got Alan Weil, who was global head of real estate at Sidley Austin. And then most recently, we hired as co-heads of the emerging company and venture capital practice at Norm Law Justin Rattigan and Sam Lipson. Justin was the general counsel of Bain Capital Ventures, and Sam was a partner at Pillsbury who did a lot of work for Bain Capital Ventures. They co-teach a VC class at Georgetown Law.
So they’re a great pair. We’re getting people like that — many of them with decades of experience at traditional firms, at top firms, or in top in-house jobs — who are really excited about what we’re doing.
Jen Leonard: Yeah, I actually think Mike was a guest at my law firm business models class at Penn years ago, so I remember he was very curious about law firm business models and how to innovate.
What does an average day look like for a partner who has moved to Norm Law, as compared to their life at a Big Law firm?
John Nay: They might walk in here — we’re all here together in One World Trade Center — and sit down with a legal engineer and talk through certain workflows that they would like to be working on building together.
So that’s a key aspect to their day.
And then a lot of it also is the more traditional working with clients and providing strategic judgment, but not on the billable hour, and instead as part of a broader relationship with the client.
That’s important because they can actually spend more time on that type of work than they could otherwise, because there are fewer teams and teams of associates to manage, and there is less work that they would necessarily need to be doing for the more traditional business development type work as well.
So it allows them to preserve more of their precious time for the strategic judgment that clients want out of the partner-level engagement.
Bridget McCormack: You said earlier that training the agents isn’t right for every legal workflow, and that makes sense to me.
Can you say more about what workflows you do use it for? Are you doing disputes, or is it all transactional work?
John Nay: Right now, the focus of the AI agents that we’re building is around real estate transactional work, real estate finance, real estate funds, private funds work, fund formation and related matters, registered funds work, and private equity, venture capital, and emerging company venture capital — so company-side and venture capital-side work.
So it’s transactional work, funds work, and also regulatory work, primarily SEC regulatory.
It’s a lot of what you would see in a full-service law firm that would serve a Blackstone, for example, or other corporate clients — primarily, right now, financial services firms here in New York.
In terms of what we’re not doing, we’re not focused on bet-the-company AI agents for bet-the-company work. That’s not what we’re currently focused on.
Jen Leonard: Can you talk a little bit about the distinction between Norm AI and Norm Law, and how they work together?
John Nay: Yeah. Norm AI is a technology company, and we serve in-house teams directly. When I say directly, I mean through technology. That’s a core part of what we do.
And then Norm AI powers much of the affiliated law firm, Norm Law, in its AI agents behind the scenes.
So as we’ve been discussing so far, all the work of the build-out of the AI agents and their deployment and their refinement — and the further embedding of client-specific information enabled by what the partners are doing with the client — that’s facilitated by Norm AI.
And then Norm Law is the law firm that is serving the client. So the attorneys who are practicing law are practicing law at Norm Law.
Legal AGI Lab, AI Agents Under Law, and Closing
Bridget McCormack: I want to hear about the Legal AGI Lab.
Tell us about the Legal AGI Lab. What problem is it solving? Why is it important to Norm AI and Norm Law? And where do you think it’s going? What are your goals for the Legal AGI Lab?
John Nay: A lot of it comes back to what I was saying earlier around agentic law and the two sides of what we’re doing there.
On the first side, working toward further automation of certain legal and regulatory tasks, a lot of that is around proprietary benchmarking work — building out applied benchmarks to really understand, in high-stakes, specific scenarios, what AI can and can’t do, and where you can really trust it to do certain work.
That part is very AI engineering and AI research focused, but with a strong overlay of lawyer understanding in specific use cases.
The second one is more around agentic law in terms of AI agents governing other AI agents, and using law as a democratically determined database of what AI should and shouldn’t do — for a form of AI alignment, is one way to think about it.
That is both AI research and AI engineering, like the first part I was just discussing, but it’s also policy and thinking through, from a traditional legal research perspective — not AI-infused legal research — how to have AI agents, as non-human entities, enter into the legal fold of what humans are in every day.
This is really important as AI becomes more agentic in the economy more broadly, not just within legal.
Having the legal analysis of how an AI agent would be treated with respect to all the laws and regulations that humans are subject to is important for society to grapple with. And it’s something that we think is relevant to our business as well.
For example, an AI agent entering into a contract, or an AI agent conducting some activity that ends up being considered criminal activity — those things, and many other things, hinge on the intentionality of the actor in that relevant situation.
So how much intention did an AI agent have? How do you measure that? And how do you see if it passes a test that a human would be put through in the legal analysis?
That’s one example of something we’re working on. I have a paper that was just posted on Stanford Law School’s website, co-authored with a Vanderbilt Law School professor, about that exact topic.
And then we do corresponding AI research on things like that, where we run a bunch of simulations, see how AI agents would respond, and see if they pass different tests for intentionality.
We’re also doing a lot of work around fiduciary duties and how well AI agents could be subject to those types of rules and standards.
For example, in the world of investment advisory, obviously, whether or not you’re a fiduciary is a very important thing to figure out if you’re going to be a registered investment adviser and provide AI-driven investment advice.
That’s something where we do a bunch of tests, and we can model it out from a technical perspective. Then we have an overlay of securities lawyers and others — ex-regulators, former commissioners with the SEC — who get involved to give oversight of that research as well.
Bridget McCormack: Is there a website people can look at if they want to check out the Legal AGI Lab?
John Nay: Yeah, it’s lab.norm.ai.
Bridget McCormack: Awesome.
I have one last question. Who’s Norm?
John Nay: The company is named after the idea of setting the norms — setting the norms for legal AI, setting the norms for AI agents being governed by law, and the other high-stakes things that we do.
And then also, a subsidiary meaning of it is “normative.”
Bridget McCormack: You know, the Gen Xers are going to think of Norm from Cheers, but you’re too young to understand that. But it’s got good connotation, so I like it.
Jen Leonard: Norm set a lot of standards too, in his day.
Bridget McCormack: That is absolutely right.
Jen Leonard: John, thank you so much for joining us. This has been super interesting, and I’m really excited to know more about AI-native firms.
And thank you to everyone out there listening on this episode of AI and the Future of Law. We look forward to seeing you on the next edition.
Until then, be well.