In this episode of AI and the Future of Law, Jen Leonard and Bridget McCormack discuss Kirkland’s AI announcement, the broader move from AI adoption to AI building, and why law firms need AI strategies that reflect their own strengths rather than copying another firm’s playbook. They also examine growing concerns among junior associates about how AI is changing early-career legal work, including questions about training, meaning, skill development, and whether traditional learning-by-doing models still work in an AI-enabled environment.
Throughout the episode, Jen and Bridget return to a central theme: as AI reshapes legal work, firms will need to be intentional not only about technology investments, but also about the people, training, and judgment that make legal services valuable.
Key Takeaways
● Big Law is moving from AI adoption to AI building: Kirkland’s investment signals a shift from law firms’ using third-party tools to building proprietary AI platforms around firm-specific knowledge.
● AI strategy should match firm strategy: Jen and Bridget emphasize that firms should not copy Kirkland unless that approach aligns with their own strengths, clients, culture, and market position.
● Junior lawyer development needs redesign: If AI can perform more entry-level tasks, firms need clearer plans for how junior lawyers will build judgment, skill, and professional confidence.
● Learning by doing is not automatic: Bridget notes that traditional training only works when someone takes responsibility for making sure the work actually produces learning.
● AI creates new advisory needs for lawyers: Jen and Bridget’s Anthropic discussion highlights how model risk, regulation, security, vendor dependence, and model shutdowns may create new areas of legal counseling.
● The future of legal work is strategic, not one-size-fits-all: Different firms may take very different approaches to AI, and the right strategy depends on what already makes a firm distinctive.
Final Thoughts
This episode places one of Big Law’s most closely watched AI announcements in a broader context. Kirkland’s investment may be significant, but Jen and Bridget caution that the real issue is not whether every firm should build its own AI platform. Instead, the more important question is how each firm will use AI to strengthen its own strategy while preparing the next generation of lawyers for a profession where the path from junior associate to senior advisor may look very different.
Transcript
Intro + AI Aha! Moments
Jen Leonard: Hi everyone, and welcome back to AI and the Future of Law, the podcast where we explore all of the changing dimensions of artificial intelligence and discuss what they might mean for the legal profession.
I’m your co-host, Jen Leonard, founder of Creative Lawyers, here with Bridget McCormack, president and CEO of the American Arbitration Association. Hi, Bridget.
Bridget McCormack: Hi, Jen. Great to see you.
Jen Leonard: Great to see you too. And I’m really excited for today’s episode.
For anybody who’s new to the podcast, we have a consistent format we use in our guest-free episodes that includes an AI Aha! from Bridget and me — something we’re using AI for that we find particularly interesting — our What Just Happened segment, where we try to connect legal professionals with the broader tech landscape and help them think through what it might mean for legal, and then a deep dive into a main topic.
And today we’re going to talk about Kirkland & Ellis’ big announcement of their $500 million AI bet and how it connects with an ongoing conversation we’ve been having about the future of junior attorney development.
So I’m really excited for today’s episode. It pulls together a lot of themes that we’ve been talking about on the podcast already.
But before we dive in, I would love to hear your AI Aha!, Bridget. What have you been using AI for lately?
Bridget McCormack: Well, I was camping last week.
You know that I always use it to identify animals by their scat. It’s a really good way to figure out what’s near you. And I also use it to identify plants and just things that I’m trying to understand.
But it turns out it’s also helpful when you’re sick and you have weird symptoms, or, for example, when you fall off your bike and you’re going too fast, and you’re trying to figure out if you actually need to go get X-rays or what the issues are.
All of the above happened to us, and I felt like Gemini and Claude were kind of like the Swiss Army knife of my vacation.
For example, my husband and I were going to ride one day in a new area that we hadn’t ridden before. I knew about how long we wanted to go for, and I knew what I wanted to see on the way and where I wanted food stops. Strava is pretty good. It has a bunch of premade routes on it.
But I actually fed some of my Strava ideas into Claude and said, “Look at these, but this is what I don’t like about them. And here’s what I’d like to do instead. Can you give me a route that meets all of my needs?”
And the answer is that it can.
Same for prescription medications — understanding them and understanding some of the details that the pharmacist didn’t have time to give you, for example. Really, really useful.
It was basically the Swiss Army knife of my vacation. That’s my AI Aha! You don’t really need to pack your Swiss Army knife anymore. Just pull out your phone and ask Claude.
Jen Leonard: Well, I’m so happy you had a Swiss Army knife. I’m so sorry that you had such a rocky vacation week.
Bridget McCormack: It was still amazing because riding bikes and being in Northern Michigan is as good as it gets.
I will tell you, when I was at the pharmacy to pick up a prescription because I did fall off my bike and was not in great shape, I checked in at the local CVS. It was like an hour after the doctor had called it in, and the pharmacist said it might be ready in 30 minutes or maybe less.
And I was like, “Okay. That leaves a lot of variation, but okay.”
And he’s like, “I don’t know what to tell you. This is not a good week to come to the pharmacy. You should not come to the pharmacy this week.”
And then the guy after me had the same thing. He checked in, and the pharmacist said the same thing: “It might be 30 minutes, might be less.” And he said, “You know, it’s just not a good week to come to the pharmacy.”
And the guy said, “But I have poison ivy today.”
Jen Leonard: The pharmacist is making me think of Office Space, and I’m just wondering if he says this every week, no matter when it is.
Bridget McCormack: One hundred percent. Best movie ever.
Jen Leonard: And there’s some coworker behind the desk that cringes every time he says it.
Bridget McCormack: What about you?
Jen Leonard: Mine was really simple, but it was one of those situations where I was just delighted by a new feature of Claude.
And I did probably the simplest use case that you can do on AI, which was I wanted to make a dessert to bring to a Fourth of July party. So I just asked Claude, “Spin up an easy-to-make dessert for a neighborhood party.”
And it did, but then it went into cooking mode.
It comes up with this little window that says “cooking mode.” Basically, you click on it, and it’s the recipe, but it actually takes you through the steps one at a time. If there’s a timed element, it starts a timer on your phone.
Bridget McCormack: Wow.
Jen Leonard: You just see the screen, and it says something like, “Cut strawberries and blueberries in a manner that you can use them as a topping.” So that’s step one.
And then you click the button, and step two is something like, “Macerate your strawberries, sprinkle sugar on them, and leave them for ten minutes.” But then you just hit play and the ten-minute timer goes on.
And then when your timer app comes up, it’s like the normal Apple iPhone timer, but it’ll say “strawberry maceration” at the top. So if you have multiple timers going, it identifies the step that you’re on in the recipe.
Bridget McCormack: That’s really cool. I’m going to try that.
Jen Leonard: They are always coming up with new things over there at Anthropic.
Bridget McCormack: They ship new things so fast.
What Just Happened? Anthropic, Model Risk, and AI Regulation
Bridget McCormack: Well, a lot has happened since we’ve had a What Just Happened segment. So I know it was hard for us to figure out what to focus on, but I think we’re focused on maybe the biggest AI story since our last What Just Happened. So tell us what just happened.
Jen Leonard: Sure.
People who are listening may have heard about the standoff between Anthropic and the U.S. government.
This is a story that we’ve been touching on a little bit at different points, which is that the federal administration had been very hands-off — and we’re going to talk about that a little later — in terms of regulating AI companies. But Anthropic seems to have become the focus of the regulations as they are emerging.
This involved the release of Claude Fable 5, which is an agentic model, and the release of Mythos 5, which is a consumer-grade agentic model.
I think that in an earlier episode, our What Just Happened was about Mythos and why it was so troubling to experts in the industry, and how Anthropic created a cybersecurity program called Glasswing in the wake of the Mythos 5 testing, which it did in-house. So it wasn’t released to the public like Mythos 5 is.
But Glasswing was an effort to bring together thoughtful people in the artificial intelligence community to test, red-team, and think about the guardrails that they would need to put up to prevent a catastrophic security incident with Mythos 5.
So Fable 5 and Mythos 5 were released to the general public on June 9th of this year, and three days later on June 12th, the Commerce Department issued an export control directive citing national security and ordering Anthropic to cut off access for any foreign national worldwide to its platform.
Because the directive was so expansive, and there was no way to restrict it only to people using it outside of the U.S., they pulled both models for everyone, which created a global shutdown across Claude apps, their APIs, and every major cloud platform.
The instigating incident for this was a reported jailbreak. Researchers, who they think are at Amazon, found a way to bypass some of the safeguards Anthropic had put into place and reach the cyber capabilities of the underlying Mythos model that had created problems earlier.
Anthropic has disputed how serious the safeguard jailbreak was and thought that it was really manageable. Nonetheless, the suspension went into place.
But after 19 days of suspension, Fable 5 went back online, and we now have access to it again. And I will say, I am in love with Fable 5. It’s amazing.
But that is what has transpired in the last month or so at Anthropic.
Bridget McCormack: I would echo your substantive use of Fable 5. I did have access to it for those three days it was out in the wild before it was pulled down. I actually had a big project that week, so I was in love with it a few weeks ago.
And when it disappeared, I was like, “Oh wow, that’s really not great because I wasn’t done with my project.”
But it isn’t clear to me why it’s okay to be back out in the wild. I do understand that the Anthropic team promised to proactively look for risk, but I think they’ve been doing that all along. I think all the labs have. I don’t mean to single anybody out.
And they promised to help with standard setting and reporting malicious activity. But again, I think those were already commitments the labs had made.
So I don’t know if it just took some time internally for Commerce and whoever else in the federal government was working on the issue to get comfortable with the model being no different from some of the other models that are out there, or if, in fact, some part of the model was walked back.
Jen Leonard: Yeah. It’s unclear what, if anything, happened during that suspension period to change the circumstances.
Bridget McCormack: And I think it’s the first time we’ve seen a model that the government has shut down, or caused to be shut down, or pulled back after its release from one of the frontier labs — or maybe any labs. I’m trying to think of any example at all.
It’s unusual because we’ve been hearing that competition requires letting the labs cook. That’s kind of the argument for a not-so-strong regulatory approach to AI: because American labs need to make sure that they’re leading the world with their products.
But there are obviously good arguments for regulating products that have safety risks. We do it in all kinds of other areas, and we’re certainly comfortable with regulation around cars and nuclear energy. We’re pretty comfortable with the idea that regulation can make products safer.
This might be the very first time there was some concern that AI might be one of those products that needs that kind of attention as well.
What do you think it means for lawyers who either advise AI companies directly or are advising clients that are using AI models? What lessons should we, as a legal profession, learn from this particular incident?
Jen Leonard: I mean, perhaps one lesson is to expect the unexpected, because things change so quickly.
Last year, as you said, the laissez-faire approach to regulation of AI companies has shifted into the government taking a firmer hand in regulating it.
But also, one of the things I’ve been thinking about with respect to the dynamics of this specific situation and the question of advising AI companies is that it seems interesting to me that Anthropic, when it realized that Mythos had these security concerns, assembled this Glasswing cybersecurity program.
To me, that seems like a helpful and intentional approach to trying to design ways to make sure that it’s not exploited in a negative way.
And to me, that was a best practice, and also something you probably would not necessarily expect from a private company trying to win in the AI arms race.
And yet their model, which I believe is fairly similar to some of the models coming out of the other labs in different respects, is the one that has been the focus of the administration.
So I worry that there’s a disincentive to draw attention to vulnerabilities and try to bring people together within the organization if the result is going to be that that level of transparency leads to disabling your frontier-grade model.
And so I think that if I were advising an AI company, maybe unfortunately, the advice would be to think carefully about how you are communicating about vulnerabilities.
I’m not sure exactly what the best strategy is, because my strategy would have been what Anthropic tried to do the first time.
I don’t know if you see a connection between their transparency and public communication and their becoming subject to these regulations. Do you think they’re connected?
Bridget McCormack: I don’t know.
I’ve heard commentators say that it was more targeted at Anthropic than Fable r. I’m not sure I buy that.
The thing that Anthropic has going for it, which gets it in trouble, is that they’re always first with new models. For a while now, they’ve just been putting new functionality on the models they currently have, and new models, really quickly, that are really powerful.
My understanding is that OpenAI has their latest model now under government — what do we call it? — in the government penalty box until it’s ready to be released.
I don’t know what’s happening at Gemini. I’m waiting for a new Gemini model.
But this may well be just the next era of how our government has to think about its role in the rollout of these models when they’re as powerful as Fable 5, and I think what’s coming from OpenAI as well.
It’s interesting too that there wasn’t agreement. Anthropic said right away that the model was no more dangerous, and that the alleged jailbreak was the kind of jailbreak that could happen in other models that were already out and being used by people regularly.
And that’s going to be hard for laypeople like me, and even lawyers, to sort through. When there’s disagreement across experts in this field, it’s going to be a hard question for any private risk committee or general counsel to decide how to weigh those competing opinions. It’s going to make it complicated.
Jen Leonard: Yeah. As you’re talking, I’m thinking through some of the other projects that we’ve both been working on. It seems like the best practice for dealing with something as shape-shifting as AI has been to create a North Star or a framework for decision-making.
And maybe here, that is the best use of an outside counsel’s time: helping prepare clients for if and when this happens.
What happens if you recognize a vulnerability? What happens if you’re put into this probationary period? How do you respond?
Because it does seem unclear at the moment.
Bridget McCormack: Yeah. I guess it’s added to the list of things where some expertise will really be needed.
It’s yet another place for lawyers to develop some expertise to be helpful to their clients, who have a business to run and can’t become experts in how to think about the risk of a particular model being pulled back, or a previous model that you’ve built some functionality on being shut down.
OpenAI is shutting down, I think, GPT-4. And if you built your products on GPT-4 and haven’t retrained them on GPT-5, you have to scramble.
In addition to that, you and I have talked a little bit about how complicated the token cost issue has become. You’ve heard examples of large companies using open-source models instead of the frontier lab models because they’re significantly less expensive, and they can host them in their own environment and therefore do some of the tasks that they want to do internally without even trusting the model’s promise of not using their data going forward.
They don’t even want to have to deal with that.
So there are a series of questions that now feel all interrelated, for which I bet there is going to be a lot of need for advice.
Yet another new area of legal practice. Good news, lawyers.
Jen Leonard: It also made me think of how we both listen to The Artificial Intelligence Show with Paul Roetzer and Mike Kaput, and they had a conversation about this topic.
They made the point that many people have been talking for years about the need for regulation of AI because of the increasing capabilities and the security risks that could emerge.
But when those security risks emerge, because there had not been a consistent approach, the response is somewhat chaotic and confusing.
And they made the point that, similar to us, they’ve been talking for a long time about the impact on jobs and work and the economy. And their fear is that when those topics become urgent, our response will be similarly chaotic and confused.
And my hope is that maybe we could have a better approach to education and workforce development for the future.
And it sort of leads into our main topic for today. We’re going to talk about Kirkland’s $500 million announcement, and then also some of the growing perspectives among junior associates and the people that supervise them about what their roles look like.
So do you want to kick us off, Bridget, on our main topic?
Kirkland’s $500 Million AI Bet and the Shift From AI Adoption to AI Building
Bridget McCormack: Yeah, definitely.
This Kirkland & Ellis story got a surprising amount of attention, and people are still talking about it.
Kirkland & Ellis announced that they are committing to spend $500 million over three to four years to build their own proprietary AI platform. I think there are other firms, by the way, that are thinking about this too. Kirkland was just the first one to make a big announcement that they were doing it.
And starting this year, starting right away, I think they said a $100 million investment in 2026. They’re funding it directly from revenue, so that means the partners are supporting it, right? Because that revenue would normally be split among the owners of the firm.
And the idea is that they’re going to build basically a Kirkland collective intelligence. They’re going to take the intelligence of all of their lawyers — I think they have about 100 partners — and their technologists, and build a platform that they will then be able to deploy for their clients.
They won’t license it. Nobody else will have access to it. And while they will continue to use some third-party tools, it will really be the crown jewel of their AI offerings, is how I understand it.
The chair of the firm made a pretty interesting statement about it. John Ballis, I think, said that a lot of the tools in the market are raising the floor for everyone, which is exactly why a firm like Kirkland, which views itself as competing at the ceiling, has to build something that’s better, bigger, stronger, more useful than just the off-the-shelf tools that many of its peers are buying and using.
But I do think it marks a bit of a shift from the industry conversation from AI adoption to AI building.
For the last few years, lawyers have been figuring out what tools make sense for them to use. We’ve had conversations on the show about the difference between using frontier models and using tools specifically built for lawyers, and whether the tools specifically built for lawyers are going to be eaten up by the frontier models before long.
But it was still always lawyers using other tools.
And this is, in a way, I think, the first story of a large firm saying, “We’re building our own proprietary AI platform. And it’s kind of the future of our business. We’re investing our revenue in this future.”
And then the second piece: there was this widely shared Reddit thread of lawyers reflecting on how AI is changing the role of junior lawyers, associates in particular.
As AI was more and more capable of doing the work that junior lawyers were doing, they were openly wondering why they’re needed at all.
And one senior associate said, for people who already have domain knowledge — so people who are senior associates or higher, partners — AI is an enhancement. And for those who don’t, it’s clearly a replacement.
That is obviously something people have been worrying about. But the Kirkland announcement shines a spotlight on it, right?
Because if we’re taking the expertise at the top of our firm, at the top of our profession, and then building this proprietary tool that our clients are going to be interested in, it feels like it’s maybe leaving the new lawyers, the junior lawyers, behind even more, unless we have another plan for how we’re going to make sure they become the senior lawyers who, together with their new big fancy tool, will be able to serve their clients.
It’s a topic you and I hear from lawyers about all the time: What is it going to mean for lawyers just starting out in the profession?
And while I still think that this is one of the most exciting times to be joining the profession, because of all the change and the new roles for lawyers — there are just brand-new roles for lawyers that didn’t exist even five years ago — I certainly understand why it feels pretty uncertain and scary for junior lawyers, at least the ones that were involved in this thread.
Jen Leonard: I will say, so many lawyers I talked to were focused on this announcement. I would argue, somewhat distracted by this announcement.
Nikki Shaver, who’s been a guest on our podcast, has made the point that the amount of money is not out of line with the proportion of investment that most Am Law 50 firms are making.
Bridget McCormack: I think that’s exactly right.
I don’t know the details of how they calculate the $500 million, but some of the lawyer and technologist time, I believe, is counted in it. The money itself is probably not that different from what some other firms are spending on their AI investments.
But I think the story that they’re building their own, and they believe that’s the path, could be concerning to smaller or midsize firms who just don’t have exactly the same horsepower to build a competitive product, if that was the way law firms went.
AI Strategy Has to Match Firm Strategy
Jen Leonard: The reason I’ve been thinking of it as a distraction, even though to your point, it makes sense to be paying attention to it as another law firm, is because — and I think this plays off John Ballis’ comment about raising the floor for everyone — the way that you and I have been talking about this with audiences we’ve presented to is that right now all of these firms are on a board game or a chess game where everybody has the same pieces.
And the question is how you’re going to play the game based on what your strengths are.
I think Kirkland is leaning into the strengths that they have, and their ability to invest that much is unique to Kirkland. Other firms are not going to be able to invest as much, and I think that’s why it’s become so much of an attention draw from other firm leaders.
But I would argue every firm should have a strategy that guides its overall go-to-market approach. What makes that firm distinctive?
And they spend so much time and money thinking through what makes them distinctive. I would say that your AI approach really needs to be married with whatever it was that made you distinctive before.
So if that is high-touch client service and access to your attorneys at any time, then that should be something that you’re using AI to actually enable more of.
If you’re a deeply cultural firm where everybody’s been there forever and you have the advantage of a stable population of experts, then use AI to figure out how to make the most out of that stability.
Or boutiques trying to become, as Jae Um calls them, bionic boutiques, and make them even more powerful in their domain.
So following Kirkland makes sense in the sense that they are leaning into their strengths — the experts that they have and the amount of money they have to invest.
But everybody has a strength to play.
Bridget McCormack: No, I think that’s exactly right.
If every single firm said, “Oh, that must be the right strategy,” most of them would fail.
I don’t think you change your firm’s general strategy because of AI. You figure out how to enhance your success with AI on what your strategy was.
I would love to ask Kirkland how they’re thinking about the junior associate issue. Maybe they have some of that $500 million set aside for a great new way to train associates. I’d love to hear what that is.
Jen Leonard: I mean, to the strategy question and different firms playing their hands differently, the contra example is Ropes & Gray, which has focused on the training aspect of it and given 400 billable hours a year to their junior associates to experiment and learn how to develop an AI mindset.
So to your point, two very well-regarded, prestigious firms taking different approaches, both of which I think make sense, but neither of which I think another firm should follow if it doesn’t align with their strategy.
Junior Attorney Development in the AI Era
Jen Leonard: One thing that we’re thinking about, Bridget, is on the Reddit thread piece and the junior associate piece.
A few juniors weighed in that they felt like their job now was rubber-stamping AI output, and that their skills are being commoditized.
And do we find meaning in the work of being a junior associate anymore?
Which caught my attention, because if you’ve spoken with anybody who’s been a junior associate in a law firm ever, you don’t necessarily find meaning in those first few years.
So is it the case that they’re actually finding less meaning, or are they under the impression that technology has made their experience less meaningful, when really it’s just a different type of rote work that you’re training on?
Bridget McCormack: Yeah. From where you and I sit, it’s probably hard for us to see the other side of it, because I know a lot of successful law firm partners who would not say their greatest job satisfaction years were when they were associates.
It’s so variable, right? Not by firm necessarily, but sometimes just by person.
If you’re working with a partner who gives you great feedback and really makes an effort to invest in your development, then learning by doing can work.
But it only works if there’s somebody who takes real agency over making sure that your doing results in your learning. It doesn’t happen on its own.
We saw recently that Hugh Carlson from Three Crowns and Megan Ma from Stanford have been working on this new AI learning tool that they have a name for. I think it’s called “Atelier”. It’s an AI simulator that actually allows associates to cross-examine or direct-examine or do other litigation tasks in real time.
It’s kind of like your simulated classes in law school, which were pretty valuable — at least valuable when you had good feedback. Again, the feedback is really important.
But being able to do that could be significantly better for learning, I think, for lots of junior associates than the previous model.
I would love to hear what other firms are doing. And maybe we’ll hear more about that as the year goes on.
Jen Leonard: Yeah, absolutely.
We’re coming to a close on our season three episodes, but I bet by the next season we’ll have some even more interesting examples of firms that are redesigning junior development.
So I’m really excited to keep following those with you and our audience.
And with that in mind, thank you to everybody out there for tuning in today to AI and the Future of Law.
We hope you’ve enjoyed this episode, and we look forward to seeing you on the next edition.
Until then, be well.