Rather than top-down mandates, these innovations emerged from individual curiosity, experimentation, and a willingness to learn. The episode also examines what “vibe coding” means for law firms, legal education, and the future of junior legal roles in an AI-driven profession.
Key Takeaways
- Agentic AI at Work: Tools like Claude Code move AI from assistant to collaborator.
- Creative Associates Lead Change: Innovation is increasingly driven bottom-up.
- Vibe Coding Lowers Barriers: Lawyers can build without deep technical expertise.
- Culture Enables Innovation: Psychological safety and time matter more than tools.
- Junior Work Is Transforming: AI reshapes early-career legal roles.
Final Thoughts
This episode underscores a fundamental shift in legal work: creativity and experimentation are becoming core professional skills. As AI handles more routine tasks, lawyers who can identify problems and build solutions will shape the future of the profession. The tools are here. The opportunity now lies in how boldly the profession chooses to use them.
Transcript
Jen Leonard: Hi everyone, and welcome back to AI and the Future of Law, the podcast where we explore all of the interesting dynamics in the world of artificial intelligence and think about what it means for the legal profession. I’m your co-host, Jen Leonard, founder of Creative Lawyers, here in person in Philadelphia, in three dimensions, with the fabulous Bridget McCormack, president and CEO of the American Arbitration Association. Hi, Bridget.
Bridget McCormack: Hi, Jen. It’s so fun to do this in person.
Jen Leonard: A treat for sure. And though you can’t see him, we’re here with our wonderful producer, Aaron Tran. And gratitude to Aaron for hooking us up at Video City here in Center City Philadelphia.
So, for those who may have listened to us before or not, we usually cover three main areas: an AI Aha!—something that each of us has been using AI for in our lives that we think is particularly interesting; What Just Happened?—something in the broader tech development landscape and what it means for lawyers; and then our main topic.
And today we’re going to talk about creative associates, which I’m really excited about because we predicted in our predictions episode that 2026 would be the year where creative lawyers really bring their talents to bear. We have a couple of cool stories about that. But let’s start with our AI Aha!’s. Bridget, do you want to kick us off? What have you been doing with AI, and what are you learning?
AI Aha! Moment
Bridget McCormack: So my AI Aha! is a little bit of an AI uh-oh, but maybe you can help me with it. You know that I often will use the same prompt across the three frontier models that I work with pretty regularly—ChatGPT, Gemini, and Claude. And you and I have been talking about how much better the frontier models have gotten at producing slide decks. In particular, Google’s Notebook LM is producing really amazing slide decks.
I did one recently that I used in a presentation in Geneva that was amazing. And I know you’ve done some for your own personal education. And my friend Vivek Sankaran, who I know listens to the pod and teaches at U of M, told me that he had really good luck with ChatGPT recently with building a deck. And I was like, oh—the last time I tried ChatGPT on this particular use case, it didn’t work for me. So I thought, time to go back to the well.
So I wanted to try building—I have lots of decks. You and I both have decks that we use parts of, like you can reuse them for different presentations. And I have a lot of those myself. I wanted to give each of the frontier models a new prompt where it had to do a little bit of research. I gave it the things that I thought it was likely to find in the research.
The topic is dispute avoidance resolution boards, which is a process that sometimes big commercial construction contracts will have. Basically, you set up a board of arbitrators or mediators—three people—who are there to address problems before they fester. If you leave disputes in the wild, they grow into really expensive, problematic, project-stopping issues. But if you address them when they’re just an annoyance, you can work through them and the project can stay on track. And for big projects, that makes a difference.
Construction has had a ton of luck with this, and it’s been around for a couple of decades, so it makes sense to me that it should work with other big projects, like life sciences. You can imagine others, but I don’t know why it hasn’t happened yet.
It also seems to me that, in particular, an AI arbitrator could be a great panel member for a dispute board, right? So there are two things I want to explore: how to put an AI arbitrator on a dispute board, and what other fields this makes sense for.
I had a long prompt laying out my thinking—my instincts, the possibility that research might prove me wrong. I asked it to do the research first, figure out a thesis statement about how AI could make a difference in this process and what the next frontiers are, and then build me a slide deck. I gave it to Claude, I gave it to Notebook LM, and then I gave it to ChatGPT.
Notebook LM produced an unbelievably intricate, beautiful deck. The slides themselves were off the charts. The content was imperfect, and I’m sure there’s a way to fix that. I probably should have worked on the content more before letting it build the deck.
ChatGPT, for the life of me, produced the crappiest deck I’ve ever seen—and I think it’s me. I went back to Vivek and texted him, like, I can’t get it to produce anything good. It won’t put in images, it won’t use any color, the words are all off the slides. I don’t know what’s going on.
Vivek sent me what it produced for him, and it was beautiful. I asked ChatGPT, why are you giving me this crappy black-and-white, words-only deck? And it said, I’m sorry, that’s all I can do. I don’t have the capability of adding images. I was like, well, you did it for my friend. Then it said my friend must have been using another interface. So I don’t know what’s going on there.
And Claude, by the way, did the best job for me. Substantively, it was really good. The slides were A-minus. But I couldn’t figure out how to change the things I wanted to change—another jagged-edge problem. Some slides had overlapping words. Some were great. The content was strong. But no experience was perfect. So—any advice?
Jen Leonard: Well, I’m wondering if you can take the content that came from Claude that was so strong and feed it to Notebook LM and ask it to make a better deck. I don’t know if that would work or not.
Bridget McCormack: It’s not a bad idea. Just take the deck and put it into Notebook LM and say, I like this content—can you produce a better deck?
Jen Leonard: I also find ChatGPT to be very temperamental. It just refuses to do things it absolutely can do.
Bridget McCormack: Like it’s in a bad mood today. Alright, you probably had better success. What’s your AI Aha!?
Jen Leonard: I did have better success. It sort of blew me away. I used Notebook LM to create a slide deck of veterinary studies because of a diagnosis our French bulldog got at Penn Vet—shout-out to Penn Vet, which is incredible.
Bridget McCormack: And also, I’m sorry about your dog.
Jen Leonard: Thank you. He’s responding well to treatment, so hopefully things will go well.
I used Notebook LM to research the diagnosis because I didn’t understand it and didn’t want to waste the doctor’s time asking a million questions. It found veterinary scholarly articles, I fed them in, and asked it to make a deck. I don’t know whether all the content is accurate, but it was consistent with everything the vet told us, and I uploaded the discharge papers as another source.
All of the protocols they gave us were reflected in the deck. And the deck itself blew my mind—the cover image was a transparent French bulldog head with the brain visible. It absolutely blew me away, and I had not used it for that before.
Bridget McCormack: I hope you bring that deck to your next vet appointment. I remember how much my pediatrician loved it when I brought in my WebMD references. If you show up with a deck that distills the literature into slides—
Jen Leonard: “I’d like to walk you through what I understand the diagnosis to be—stop me anytime, I can take questions as we go. I probably have more questions for you.” That would be amazing.
I was also playing around with Notebook LM, but my bigger AI Aha! moment this week connects to our What Just Happened?—Claude Code. I’ve wanted to revamp our company’s website for a while, but I lack the skills, time, and patience. We were listening to Hard Fork, and the hosts were talking about how they used Claude Code to build their own websites. That inspired me to try it for my own website.
I spent about three hours with Claude Code and couldn’t believe how beautiful it was. I put in our existing URL, then shared a few websites I liked, and they were all different. Even though they weren’t similar, Claude analyzed them and identified the unifying themes that were drawing my attention. I also mentioned the aspects I don’t love about our website, and it started coding immediately—going bonkers.
I feel like Penny from Inspector Gadget. I always wanted that computer book, and now Claude Code is just going, and I feel productive even though I’m not doing anything.
It generated all the files I need to hand off to a web developer. I can’t activate the site myself, but I have all the HTML files. It also generated a README for the developer with specs for fonts, colors, and two unresolved issues—our team photos were embedded in the HTML and not centered properly. The README basically says, here are the things you still need to fix.
The site is so much more beautiful than anything I could have come up with. I did laugh because the color palette looks just like Claude—terracotta, cream, gray. Claude said, “This seems very sophisticated and bespoke.”
One thing people mention with Claude Code is that long conversations can degrade. That happened toward the end. It started rejecting new prompts, probably because of the context window. I also had to upgrade mid-session to a higher-tier subscription because it said I had nine prompts left. I upgraded to the $100-a-month plan from the $20 one—and honestly, it was unbelievable.
What Just Happened
Bridget McCormack: That’s a great lead-in to What Just Happened?—because our What Just Happened? is Claude Code, and this latest version of it has taken the internet by storm, especially over the holidays when most people were trying to stay off it. A lot of the people we follow on LinkedIn and other social media sites were remarking on what an astonishing release this is.
You now know, because you’ve actually been working with it, and it sounds like you’ve had a similar experience to the people we follow. Can you tell us a little bit more about what this latest Claude Code does? What makes it different from what it could do two months ago? Why do we care? Why does everybody seem to care so much?
Jen Leonard: I’m happy we’re doing this because I was one of those people who completely unplugged for the holidays. I’m not on social media platforms where people are talking about this a lot, so I was sort of caught off guard when I got back from vacation that everybody was talking about this. It’s one of those buzzy things that just ends up in the ether—I don’t even know where I heard about it first, but I just started hearing about it everywhere.
So Claude Code is a tool that essentially allows developers to work directly with Claude from a command-line-type interface, which is familiar to coders—not familiar to me. Instead of having to do all of the copying and pasting that coders usually do back and forth with Claude to get it to work, this is a much more agentic system where Claude is working alongside them inside their actual coding environment. It breaks down a huge friction point.
Claude described it as the difference between texting your contractor pictures of what you’d like them to do in your kitchen versus having your contractor standing in your kitchen, looking all around without you having to be the intermediary to give them information. So people are excited about it.
We follow Andrej Karpathy pretty closely—an incredible engineer, computer scientist, programmer, and co-founder of OpenAI—and he said he’s never felt this far behind as a coder.
His quote was, “I’ve never felt this behind as a programmer. The coding profession is being dramatically refactored, as the bits contributed by the programmer are increasingly sparse in between. I have a sense that I could be ten times more powerful if I just properly string together what’s become available over the last year. And a failure to claim that boost feels decidedly like a skill issue.”
So if you feel behind, don’t feel badly—because Andrej Karpathy feels behind.
It’s more agentic in the sense that Claude can take action on its own rather than just give advice. You can leave it alone, come back in a few minutes or an hour, and it’s done all sorts of things. It understands context extremely well because it can see the entire code base it’s working on, so its suggestions fit the whole project rather than just a snippet.
Apparently this is a really big deal because most AI coding tools only see what you paste in, not the broader context. Claude Code also handles a lot of tedious but time-consuming work—updating documentation, refactoring code, setting up boilerplate—so developers can focus on higher-level strategy. That should sound familiar to lawyers.
Because it was such a game changer for coders, Claude has now made it more accessible. I couldn’t really use it well in the desktop version—it was confusing because I’m not a coder—but now you can access it right from your browser. They also have Claude Co-Work, which is sort of the evolution for knowledge workers. It’s an incredibly powerful glimpse at what it will be like to work with an AI agent who can complete tasks without additional prompting.
Bridget McCormack: This feels like the ChatGPT moment, but for coding. Even people who were following generative AI closely were surprised by how big of a step up this was.
Jen Leonard: I don’t have coding experience, and I don’t really understand what coders do day to day, but watching Claude generate all this code made me think: I can’t believe this is what people used to do for their jobs. Just typing in all of these symbols for hours—like The Girl with the Dragon Tattoo, staying up all night understanding this complex coding language.
It made me think about law, too. People might look back and ask whether junior associates really used to sit in rooms with bankers boxes, manually reviewing documents, sorting them into piles, and creating privilege logs. It sounds crazy now, but that’s what we used to do.
So when I look at coding, I think about how much more satisfying it might become as a career—and how transformative this is.
Bridget McCormack: Do you have thoughts about what it means for lawyers? Does the Claude Code moment mean we’re going to have Claude Legal in six months? And what does that look like?
Jen Leonard: I know we’re going to talk about this more in a future episode, and you’ve thought about this more than I have. Yes—and I’m not really sure what all of this agentic behavior means for the supervising lawyer, because the system is making decisions on its own. You surfaced some additional questions about what it means for contract law itself—when an agent is making decisions, it’s no longer just a meeting of human minds.
But in terms of how the work gets done, absolutely. I could imagine—maybe it’s already happening—if you’re a litigator in discovery on a small case, dropping all the documents into Claude, explaining what you’re looking for and the context of the engagement, and running eDiscovery far more efficiently.
I don’t see how it wouldn’t disrupt junior-level work. What do you think?
Bridget McCormack: I think we have reason to believe the frontier labs are working with lawyers the same way they worked with doctors to produce ChatGPT Health, which came out recently. They had doctors do reinforcement learning on frontier models to improve health-related responses.
I’ve heard researchers say they built coding capacity first because that’s what they know. It was useful to them initially. If it can code autonomously, they can supervise the AI agents so humans can focus on higher-level strategic thinking—deciding what questions matter and what problems are worth solving. We’ll be able to create AI researchers who are going to make the breakthrough. They’ve also said that coding isn’t where this ends; it’s where it starts.
So I assume Claude Code for legal is coming, and that they’re already working on it. There are a lot of things lawyers do that they’d be happy not to do, so they can focus on work that’s more meaningful and valuable.
Jen Leonard: In academia, especially scientific research, there’s been a lot of discussion about how accelerating discovery doesn’t necessarily accelerate execution because of bottlenecks like peer review.
Bridget McCormack: Those human gating moments slow everything down.
Jen Leonard: I’ve been thinking about that in the context of Claude Code—what can be accelerated, and what can’t. Even if you can’t speed up something like a clinical trial, are there ways to accelerate peer review itself?
It gives me hope that bottlenecks won’t always prevent breakthroughs, which is one of the things I find most exciting about AI. That leads nicely into talking about creative lawyers—people who are identifying problems and experimenting with AI to build real tools. It makes me optimistic not just about improving practice, but about expanding career paths and letting lawyers exercise skills they haven’t always been able to use.
Main Topic: How Creative Lawyers Are Using AI Tools
Jen Leonard: So we found two really cool stories from the same firm. I don’t know what they put in the coffee over at Linklaters. But their associates are super creative and ambitious. So do you want to share what happened at Linklaters?
Bridget McCormack: Yeah, it’s very cool stories, in part because it seems like each of these associates, just on their own, created these AI tools to solve annoying problems in their day-to-day practice.
So it wasn’t like Linklaters said, “Here’s an AI training, and here’s your assignment. You’re going to have to do this.”
The first is a managing associate in banking and finance in the London office. Her name is Tanya Sadoughi. She was trying to solve the problem of clients and partners wanting more frequent updates on the work in progress, especially with respect to fees.
They have that information, but producing it was pretty manual and therefore tedious and time-consuming. And not a lot of value is created by producing the update. I mean, the client wants the update, but the work of producing it isn’t really valuable—and it’s obviously not that fun.
So she built a tool that just generates those updates automatically from time records. You have a button, you press the button, and you get a client-ready summary. That’s really valuable. The client wants that information, and the partner does too.
She apparently worked on a Python prototype over one weekend in July, pitched it to her leadership the next month, and by the next year everybody across the firm was using her solution globally—like, everybody was using it.
According to her, she just taught herself how to code on weekends, in part because she believed she was going to need to upskill pretty aggressively. She believed that after ChatGPT launched, the profession was changing. Sound familiar?
So it was worth figuring out how she could learn new skills and bring them to bear in her legal practice.
That was pretty exciting. I love the self-starter-ness of it, and it makes me excited for lawyers coming out of law school right now, because I think they’ll be fearless about learning these new skills and figuring out how to do things totally differently.
Students and brand-new lawyers are always the ones who come up with breakthrough solutions. Even before we had AI or even tech, it was always a student—especially in clinical teaching—who would look at a statute and say, “I think we can file a motion to do X.”
And I’d be like, “Oh no, we don’t do it that way. We never do it that way. It’s never been done that way. And they’d be like, “Well, what if we just tried?” And I’d say, “All right, let’s try it out. Because it’s a clinical class—we can try. The judge can tell you no.”
And every once in a while, one of those would work. And I’d be like—right, it’s always better not to be used to 20 years of doing it the other way. It feels to me like that on steroids.
Jen Leonard: I just love Tanya so much. I’ve never met her. But you said this in a presentation we did recently—it doesn’t really matter how we feel about things. This is the state of things that we need to respond to.
And I feel like in our profession in particular, there’s such a culture of not moving quickly to that place, but instead sitting in our feelings a little bit about how everything is and how it should be better, and how it should always be the way we learned how to do it.
Not for Tanya. She’s like, “If nobody’s going to teach me, I’m going to teach myself Python,” even though she’s a Big Law associate and probably has other things she could be doing with her time. She’s recognizing how the world is changing and taking it upon herself to change with it.
I don’t know how you teach yourself Python on the weekends, but I just thought that was incredible.
Bridget McCormack: I actually think with AI, though, you could. I mean, that’s why I teach myself all kinds of things that I never would have tried to teach myself before.
I love your reminder that it doesn’t matter how we’re feeling about things. When I was running for office in 2012, it was Fourth of July weekend, and we were going to do something like 20 parades. We did one in the morning, it was a million degrees, and we were supposed to go do another one.
My sister was in town with her kids, and she sat on the floor and was cranky because it was so hot. She said, “I don’t want to do it. I don’t want to go.” And my sister was like, “Yeah, I don’t either. Nobody does. None of us want to do it. That’s what we’re doing right now.”
I don’t want to do it either—and that’s what we’re doing right now. We still say that all the time.
All right—back to our story. I love James Phoenix as well. He’s a managing associate in litigation and investigations. He was frustrated with mistakes they were seeing in their time-recording narratives, because that meant they had to do this really tedious manual review before they could actually send bills out—which makes sense. You can’t send the bills out with mistaken narratives.
So he built an Excel plug-in that uses their internal chatbot—apparently they call her Leila—to flag those issues and fix common narrative errors, things they see over and over again.
He vibe-coded a little Excel plug-in that uses another internal AI tool of theirs to do this, and that’s now rolled out globally across the firm. And that’s also amazing.
It’s like, “This thing is really annoying. I’m getting no pleasure out of it. And now we have a solution.”
So I’m going to go figure out how to build that solution, because apparently anybody can vibe-code now—anybody but me.
What I love about both of these stories is that they took it on their own initiative. It wasn’t some firm strategy, although there must be some firm innovation culture—that’s my guess.
But I love both of these examples, and they give me a lot of hope about the future of up-and-coming lawyers, who seem pretty innovative.
I’m trying to think about what lessons we should draw from this—for law firms, for training, for law schools. James used the term “vibe coding” when he was describing it.
That’s the idea that you can just tell AI what you want and it will help you build it. You even just did that with your website. You don’t need deep technical skill—I don’t believe you do, and James said he doesn’t either.
So it seems like it might lower the bar for lawyer innovators, which feels hopeful. Is that a takeaway you feel?
Jen Leonard: Yeah, I definitely think it lowers the bar. I think a couple of things need to happen to really grow this movement. It strikes me that Tanya and James, and some of the students and young lawyers and junior lawyers we know, fall into this bucket of people with an entrepreneurial mindset. They’re looking for problems, and they’re not waiting for permission to solve them.
Most lawyers, to overgeneralize, are not temperamentally suited to that. They tend to be rules-followers. They’re excellent at what they do, but working in a law firm can be very scary if you go outside the norm of what the firm expects.
I remember a mutual friend of ours talking about how, when he was a litigator in the ’90s, he used Control-F and told the partner, “We could do this so much faster.” And the partner was like, “Don’t do that. This is how we do it,” because it impacts the billing model and feels threatening.
So I think creating a culture where this kind of experimentation is celebrated and rewarded is step one. The technical capability is already there. Any associate could be doing what James and Tanya did.
I saw on LinkedIn—and I sent it to you—that Ethan Mollick is teaching a vibe founding class. He talked about shrinking an entire semester into basically four days and all of the insights students were able to generate.
I could imagine a vibe lawyering class, where you bring in students and either share major problems from the profession and have them tackle them, or they come with their own problems, find a problem, vibe-lawyer around it, and share what they learn.
I think that could be the coolest class. I suspect a lot of faculty wouldn’t be excited about a class called vibe lawyering, but that’s the point about culture. If you don’t create conditions where people feel psychologically safe and encouraged to experiment, then innovation only happens because of extraordinary people taking it on themselves.
Bridget McCormack: I think vibe lawyering, not only within a law school but maybe in liberal arts as well. In addition to law students, teach it to non-law students. Send them to the debt-collection docket at the local court and then send them off to vibe-lawyer for four days and see what they come up with. It’d be amazing.
Jen Leonard: All of these firms do extensive feedback surveys with their clients. They have years of data on client pain points. You could do vibe lawyering with junior associates or summer associates, especially when they don’t yet have a billable hour requirement.
These are great examples—especially work-in-progress updates, which are things clients want. It’s not in anyone’s best interest for people to be spending their time on this. That’s what you want to find. Hand that to your juniors and say, “Go at it. Help us figure this out.” I think that could be so cool.
Bridget McCormack: I think it’s a great idea for a summer program and another way to create some excitement and buzz about your firm.
Jen Leonard: And different from things firms have always done. People do great tried-and-true things, but this feels really cool to me.
So I definitely think it lowers the bar—not having to do what Tanya did and teach yourself Python, but being able to dive right into it.
As we talked about culture and support, Linklaters in this case—based on the articles we read—gave both associates resources. They gave them teams and budgets.
I thought it was really interesting that in both cases these associates were seconded to their innovation teams. And in Tanya’s case, the way I read it was that she was able to blend her experience working on these tech projects with practice, which in a lot of firms requires choosing a single path.
And that’s one of the things that repels people from working in firms. It doesn’t sound like that’s what happened here. So what do you think an organization can do to support these efforts, beyond culture?
Bridget McCormack: Culture feels like the foundational requirement, right? If you have a culture that creates space to learn and encourages learning, you need everybody in. I really agree with lawyers who don’t like having to choose a path, and I think that’s still fairly common—choosing between the AI or tech side, operations, or substantive legal work.
I think that’s a huge mistake. It causes people to miss opportunities to see problems and solutions at the intersections, because everyone just sees what’s in their lane.
I think you really need everybody in. I’ve said this a million times—it has to be top-down and bottom-up. Whoever the managing partner is needs to be vibe-coding herself, showing where she’s succeeding and where she’s not, and getting people excited about it.
You also have to give people time and space to actually do it, not just say you want everyone doing it. Firms have ways to structure expectations, and that’s an advantage they have—but you still need real resources. R&D is hard for partnerships, and that requires thought. Innovation happens when it’s structured. All of those pieces together lead to success.
I don’t know enough about Linklaters, but it makes me really curious about them.
Jen Leonard: I know Linklaters is a Magic Circle firm. My sense of Magic Circle firms is that they’re very established UK based firms that are traditional, although the articles did talk about how Linklaters already has a very innovative culture. My takeaway is that if a Magic Circle firm can promote innovation this way, then anyone can build a culture for change.
You’re much more of an optimist than I usually am. You’ve talked about how there’s never been a better time to become a lawyer, and these are great illustrations of why. I think we could never have done this in our junior lawyer days because the tech just wasn’t there.
Bridget McCormack: No, this really is the best timeline—even for lawyers.
Jen Leonard: Yeah. So on that note, I think we’ll wrap up our conversation on Claude Code. I can’t wait to go home and vibe-code some more, because I feel like I have superpowers now, which is really cool.
So thank you to everybody for joining us on this episode of AI and the Future of Law. We look forward to seeing you next time. Take care.