Legal Tech StartUp Focus Podcast
The Legal Tech StartUp Focus Podcast covers the startups that develop and sell legal tech products and services. Through interviews with legal tech startup founders, investors, customers and others with an interest in this startup sector, the podcast's host, Charlie Uniman, and his guests will discuss such topics as startup management and startup life, startup investing, marketing and sales, pricing and revenue models and the factors that affect how customers purchase legal tech. In short, the Legal Tech Startup Focus Podcast will focus on just what it takes for legal tech startups to succeed.
Legal Tech StartUp Focus Podcast
How Flatiron Law Group Builds a New-Model Practice With AI (and How It All Started with a 25-Pound Laptop)
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
If you think AI will replace deal lawyers, this conversation will change your mind, and it might change how you choose counsel. I sit down with Conrad Everhard and Leonard “Lenny” Nuara from Flatiron Law Group (https://flatironlaw.ai), two former Big Law partners who built a new-model boutique designed for high-end M&A, private equity, venture capital, and technology commercialization without the traditional billable-hour leverage game.
We get specific about what makes fixed-fee legal services work at the top of the market: a different labor model, a distributed firm structure, and proprietary legal technology that turns diligence into clean, reusable deal data. Lenny walks through Deal Driver, their complex transaction management platform, including clause-level extraction, narrowly scoped AI agents, and “trust but verify” reports that link straight back to source documents. Conrad explains why this is more than efficiency; it becomes a competitive moat that can improve both pricing and quality.
We also dig into training and the future of the junior associate. Deal Mentor, built with Stanford Law School, uses AI-driven negotiation simulation to give young lawyers realistic practice without risking a live transaction. And we close with a warning every legal tech buyer should hear: the monoculture trap, where legal AI regresses to the mean, often ignores things like the client's leverage (or lack thereof), and produces markups that look smart but can blow up a deal.
Subscribe for more candid conversations on legal tech, AI in law firms, and the business models reshaping legal services. If you found this useful, share it and leave a review.
Welcome And Why These Lawyers
SPEAKER_01Welcome again, everyone. This is Charlie Uniman, your host at the Legal Tech Startup Focus Podcast. And um while we often talk to founders in Legal Tech, uh this time we're going to talk to real live practicing lawyers, lawyers who do um deals and other work at the top of their practices, located principally here in the East Coast, a little uh West Coast action in Palo Alto. Uh I'm gonna have um Conrad Everhart and Leonard Lenny Nuara of the Flatiron Law Group uh join us. You'll hear what makes his or their, I should say, practices special, why they're of particular interest to people who follow legal tech. And um a big welcome to you both, Conrad and Lenny. Thank you for joining me. Thanks for having us, Charlie. Pleasure is all mine. So we're gonna begin. I've I've had the benefit of meeting these two fellows in real life out at a conference recently in California and talking to them beforehand about what we're gonna discuss. So I know them, I enjoy their company, I enjoy what they have to say about the law.
Leaving Big Law To Rebuild
SPEAKER_01Tell us both of you um how you got to where you are now with Flatiron Law Group. You were in some big law law practices. Uh, what got you to where you are? Either of you can start, go right ahead. Go ahead, Conrad. You can start and I'll follow.
SPEAKER_00You know, so so Charlie, we are former big law partners. And without without and and all of us, all of us founders at Flatiron, we all had ambition to change the model and to and to tweak the model, and we weren't getting anywhere in big law. And we found ourselves about seven, eight years ago, we found ourselves displaced from big law. There's a backstory to that. And uh um, and and and as I as I mentioned, we had ambition to to experiment. So we decided to get outside the firewall, like we always say, and start our own law firm and where we would have freedom and model, complete freedom and resources, um, put in our own capital, and uh we formed what is now called flat iron.
Fixed Fees Change Everything
SPEAKER_00Um we started experimenting with with different pricing models. Now we do principally, you know, high uh middle market MA, private equity, venture capital, deal work, technology commercialization. That's still what we do today. Um I mean, we're still a boutique, and we are, I guess what we would call now a new model firm, but we've been a new model firm for a long time. Um not so new. And not so new anymore. But the the key is the key, Charlie, is when we we started experimenting with fixed fee pricing, and we do fixed fee pricing today. And when we when we ex when when we started looking to fixed fee pricing, it caused us to rethink everything about the delivery of legal substances because we weren't now making money anymore, leveraging junior labor. We were making money like other businesses by by by expanding margin. So, you know, we've employed a lot of strategies, including the way we deploy labor. We deploy labor very differently, we acquire labor differently, we get different grades of labor, we apply it, you know, differently. Um, obviously, overhead, you know, we've gone to a to a principally distributed model, and uh and you know, we use SaaS applications to run our practices. Um, but the third part of it is that you know, we started dabbling in technology at the outset, it was third-party technology, but now that's become proprietary technology. And Lenny, why don't you talk a little bit? You're the architect of our of our technology infrastructure, you're a you're a developer yourself.
A Lawyer Who Builds Software
SPEAKER_02And so, Charlie, you know, I'm in in the first instance, uh very much uh happy to be here, but uh I am uh somewhat of a different breed with it comes to being a lawyer. I'm practicing loft for decades, but I also have a degree in computer science. And so I was always building technology uh to help my practice uh through all the years that I've practiced. Now, you know, uh Conrad laughs about this, but I I carried around a computer in law school from 1981 to 84, which you saw when we were out at the computer.
SPEAKER_01Yeah, I'll uh I'll have to say it was uh it was a whopper. It wasn't your slim spelt notebook computer. Um but uh yeah, he did carry it around, and I saw it in real life. And you know, Lenny's uh a big strong guy, and I think he began to acquire his musculature by carrying that damn thing around in law school. That's exactly right.
SPEAKER_02It's 25 pounds. Most most people had the belief that it was a sewing machine. Uh it was a portable sewing machine, but it was not. Um, and I typed my notes on it in school. And um, in any event, uh I tried to leverage the technology because I enjoyed my my time at Boston College. I was a teaching assistant in the computer science department. And then I went to law school that for the first course at Seton Hall Law School that taught uh computer law, Jetson Jennings. Uh and uh uh that that's been part of my practice for all of my years. And when Conrad and I um uh met many years ago during Elevator Pitch Olympics in New Jersey, and then we became friends together at Greenberg Trarrig and then decided to uh just to create what is now flat iron, we've always been leveraging tech um wherever we could. Um but there were certain things that we wanted to accomplish uh with doing, as Conrad suggested, flat fees and uh um and leveraging the technology in a different process methodology. So it's not just tech, it's uh a different mindset where we could essentially offer the flat fees, but at the quality that we used to uh deliver at Thatcher Profit, where I used to work, and at uh Greenberg Traveling, where uh Conrad and I both worked, as well as uh at the other firms that Conrad and I both worked at. Um so quality is is essentially the most important thing, and talent is the uh you know part of that. But then how can we re-engineer things differently to deliver the services at a really high level? And most recently, the advent of AI, and and we can talk about our work with Stanford Law School on the tools that we've built with them. Uh, but the advent of AI is um basically, you know, uh a tsunami of uh of technology rolled into an new interface that is extraordinary in the right hands. And so we're we've been leveraging that now for you know since ChatGPT came out. Um and so that's a long way of describing we have technology just not for the sake of technology, but as a core part of our practice every day. Uh, and what we have the benefit of me with the technical expertise that I can use that technology in ways maybe that weren't originally anticipated, uh, and then we can grow our practice and and deliver better services.
SPEAKER_01Indeed. And and uh, you know, we're gonna cover this a bit later, as you said. Technology in the right hands, having human hands in the picture is, as I think we've come to realize, um essential. But not just in the picture. I've heard this from these two fellows, and and and they live and breathe it, having not just the human in the loop, but I'll let them elaborate on it in a bit, a human-first approach to technology. And if you're gonna do uh top of your practice work uh and still leverage technology to be able to work on a uh fixed fee basis, you're gonna have to meld the human and the tech in the right way. And I'm gonna have to uh say that we're gonna spend part of our time figuring out just how Conrad and and Lenny see that is the right what is the right way to meld the human first approach and the technology that has to be applied in order to uh uh take advantage of the inevitable cost savings and efficiencies that that technology affords. But
Deal Driver And The Deal Workflow
SPEAKER_01you guys, along with Lenny, uh running the uh running point on this, have actually built two apps that are essential to your practice, home-built. One is, and and they're both uh uh I guess uh uh deal-driven. One is called Deal Driver, uh, one is called Deal Mentor. What is what is Deal Driver and how do you use it and how central is it to what you do every day?
SPEAKER_02Okay, so uh Deal Driver is a complex transaction management platform. Uh it it provides uh a dual architecture for both buy side and sell side, having private venues. It takes a deal from the very beginning, right after the letter of intent, uh, and manages all of the data that's necessary in a deal, all the due diligence, all the due diligence questions, uh, categorizes, organizes, and then and then after uh uh that in terms of tracking and so on, without using the bad word in uh the legal space, uh project management or task management. No, right, no lawyer, no lawyer or deal person wants to be project managed or task managed. So we kind of hide that, uh, but nonetheless, we extract data from all the due diligence requests to track them and all the due diligence responses and the follow-ups on all of them, and then we move further, tracking uh intelligence or data from the documents, from the due diligence, from the answers, um, and then use that in the perfor in the in the creation of the documents that are necessary for the lawyers. So in responding to uh specific requests, you can use that data generating reports, for example, a red flag report, a report regarding the quality of the due diligence that's that's present or might be missing, but all the way down into uh responding to representations and warranties, preparing uh the disclosure schedules, uh, preparing the closing checklists all the way through to close, post-close integration. So it's a, and I we say complex transaction management platform because it's a deal-oriented platform, but the type of deal will vary. So we do principally MA, high-end MA. Our adversaries are the big firms we came out of and all these other firms that are in the top uh 50 or top 100 law firms uh around the country or actually internationally. Um, but if someone does uh you know complex project uh project management or project finance, excuse me, project finance deals, or someone's doing uh um venture or emerging growth companies, or someone's doing commercial leasing or real estate development. All those have a similar journey, not although not identical. But they have data that they need to uh organize. They have data that they need to extract to answer questions and to put together the schedules or otherwise. And so we don't call it just a due diligence platform for MA deals. It's a complex transaction management platform. And it is critical to our uh our operations because it allows us to leverage the knowledge in the documents wherever they are, whether those documents are requests or responses and the documents themselves, the clauses, you know, we can find various clauses and so on. And we and we marshal that through the deal to make us um A more accurate. Um and obviously we always have a human, uh we call it HI and AI. We use human intelligence driving the AI. We don't just stick a lawyer in the loop or stick a paralegal in the loop. They drive the technology, and that's a that's a critical element of our architecture of the of uh our technology and the architecture of our process flow so that we can be assured um uh you know every aspect of our the quality of work. And it's one example. When reports are generated, and this is not uncommon, but it's something that has been critical to us from the very beginning of using this, we provide links, hot links that go back to the source. So if you're saying that there's an assignment provision that will somehow require 90 days of consent, it'll flag that, but it'll show you when you when click the link and it brings you right to the document, right to the place. So that way you can do that constant verification, even in a report. But ordinarily, that's checked much earlier on. If the if if the reports were are made just on the documents themselves, that doesn't always work. We break the documents down into pieces early on, and then we have those pieces verified by humans because they're asking for certain things. So they're saying, get me this, and then we're verifying that. And then later on, we can use that data over and over and over again with an extremely high reliability rate. Um, and then yet we still don't still don't trust. It's trust but verify. So we trust, we write the report, and then we have the verifiable links to just check again um before we submit something to uh to a client or or an adversary.
SPEAKER_01So that's deal, that's all deal driver. And and and when you talk about uh breaking the documents uh into into bits, is that done um at the outset by a human, or is the machine have some say in how the documents get broken down so that you can be sure the the AI is is not getting infused and overfed? What how does that breaking down into bits work?
SPEAKER_02What happens
Agents, Clause Extraction, Trust Verify
SPEAKER_02is that we have what are called agents. Everybody knows that phrase today. Um, but the agents work one at a time in uh in parallel, but one at a time. Um and they don't they don't necessarily work work autonomously on making decisions. Agents have a um similar to a let's say we'll say a young associate or a young paralegal, um, they're given limited tasks, and then we we get the uh the response from them and they come back. So we break things down into usually at the clause level, and we're extracting, you know, all of the clauses, all of the elements, all the facts that we are uh think are necessary to that deal. Uh the the platform comes with a series of things that you would ordinarily expect in MA, um, but uh uh such as assignment clauses, change and control clauses, the value of the contract, who the parties are, all that necessary stuff. But we go even deeper than that. And and uh on the platform, when unusual things uh come up, we can generate new agents on the fly. Uh we'll test it on something uh to make sure it's scoped correctly, and then we'll go extract those pieces. What we found is that the more the more things uh you jam into one prompt or one agent, uh the lower the quality is. So we we break things down uh significantly to small. So uh, you know, uh aim small um and and you'll get a much better result than aim big.
SPEAKER_01And and Conrad, how uh how how uh uh how would you describe how essential uh tooling such as deal driver is to your successfully um uh working with a fixed fee arrangement and yet maintaining the margins that enable you to buy your shoes?
SPEAKER_00It's it's uh it's it's our moat, Charlie. I mean, we have we have several moats, but it's our most important moat. I mean, we're competing against big law for premium quality transactional work. So in the in the first instance, we have to match them or even maybe even exceed them on quality. But the value proposition for the client is that we can match them on quality, but we can significantly outprice the market because we're more we're more efficient driven by our technology and our labor model. Um the other thing that was, you know, so so in in the first instance, we're allowed that we're we're we're it enables us to price ourselves on a fixed fee basis, and not it enables us to price ourselves more competitively against the market. But but interestingly, and this is not something we expected at the at the outset, um it actually creates better quality data, which is something that's very that the buyers are very sensitive to. Because we break the data down into pieces and we synchrono we tag the data and we can synchronize it with the various due diligence requests, the disclosure schedules, it surface, it cleans the data, it flows better into the silos of multi of the of the uh serial buyers like private equity funds, because they can now synchronize their data. And it also you know it surfaces issues earlier. So what we find is that private equity funds like the way we're we're handling the data, and they would they'll refer targets to us, particularly their sloppy targets who haven't been professionalized. Um and and that was the you know, that was that was the surprise in all this, is that is that the quality of the data we were producing became also a competitive differentiator.
SPEAKER_01You know, I I have long held long, you know, in the age of AI, long is four days, but I have long held that not only will the use of AI, proper use of AI, enable you to be more efficient and maintain margins and at the same time um uh price more competitively. But if you're not using these tools, we're gonna see a day soon where whether it's litigation, whether it's MA, whether it's private equity, whether it's real estate, trust and estates, I don't care. If we haven't seen it already, if you're not using it, not only are you gonna be at a competitive disadvantage outside of the price realm in terms of quality and reliability, but you're also gonna be more at risk for doing a poor job and, God forbid, being sued for malpractice.
Talent First AI Powered Strategy
SPEAKER_00Uh look, you know what the Charlie, at the at the high end of the value curve, or we are. I mean, there's there's lots of things going on in the legal profession right now. There's there there's private equity that's buying out smaller law firms. Those are just basically cash flow plays. You've got AI first law firms that that really are focused on repeatable tasks like MSAs and NDAs. Um and you know, that they're they're focused on law as as software. But it you know, we're we're sort of in the middle of all that. We're we're we're at the top of the food chain. We're doing, we're doing, we're doing M ⁇ A and high-end transactional work. There, you gotta really blend. You know, it's a we we say we say talent first, AI powered. You still need that talent if you want to compete with big law. You still need the talent for judgment, you still need the talent for nuance, for, you know, for decision making and for and for backstopping. Um, but you can be extremely efficient by uh you know accelerating using technology and AI. And that's the balance that we're we're you know, that we're seeking. Um, I don't think you can do what we do with AI alone. There's just too many downsides. But but but you absolutely need AI to accelerate the task and and to make you more efficient and more competitive on a pricing basis.
SPEAKER_01Now you juxtapose Flatiron's approach and the nature of its practice very well with, for example, uh Crosby, Norm AI, and the like. I don't know if I have each of them, I don't mean to disparage what they're doing. No, uh, and I may not categorize each of those and others properly. But and it and it may be a play that will develop differently, you know, Clayton Christensen come up from the bottom, disrupt, pick the easy stuff, and they get more complex. But right now, as I hear it, Flatiron is doing top of the market work, as you've said several times, against the big law law firms that are handling bet the company matters. While juxtapose with those that are AI first that are handling more routine, more standardized, more I hate to say, bottom of the bar.
SPEAKER_00It's a different, it's a different business model.
SPEAKER_01It's a different business and a different market.
SPEAKER_00Yeah, exactly. They're they're they're looking to be uh either an ALSP or or software. They're going to they're they're using AI to do repeatable tasks, you know, and and there's nothing wrong with that. It's high volume work. Not at all. Not at all. It's a classic venture-backed model. They're doing what whether or not there's enough of that kind of volume work to support their valuations, and and whether in fact you need a lawyer at the end of the day to do to do that kind of work, or whether the the in-house lawyers can just go to AI themselves. That's a that will be decided in the future. But but we're, you know, we're we're focused on lower volume, higher margin, higher priced work, which is which is a hybrid, hybrid model, which also is a very interesting model. And you know, normai is sort of there because they're doing that on the regulatory side. Um, and there are, you know, there's a guy from Kirkland and Alice at Irving who's who's we're not I'm not exactly sure what he's doing, but it sounds like he's doing something similar. Um, but yeah, we're we're we're we we we're we're forging our own path here.
Deal Mentor And Simulation Training
SPEAKER_01And then let's take a few minutes to talk about Deal Mentor. Um Lenny, Conrad, either of you, both of you. What is that all about?
SPEAKER_02Yeah, Deal Mentor is a uh a negotiation simulation tool. Um it's actually more than negotiation now, but it was built initially as a negotiation simulator to train uh young lawyers, uh both in law school and for their first five years, let's say, of practice, um utilizing AI to create a series of personas that the uh the mentee, the the student of the AI that would negotiate with. Um you you it comes with a certain uh series of modules, but you can build additional modules. We built this tool with uh Stanford Law School, with Dr. Megan Ma um and and and others, uh developers uh um uh and And the the purpose of it was um we were filling a gap that is generally missing at the time that we started this, which is now I think three or four years ago, um, which is a training tool that's actually engaging a true simulator that puts the students or the mentee, as we call it, um, in the chair of having to negotiate certain terms. Uh it starts, uh, the very simplest module is the the the liability protocol for representations and warranties, uh, what's going to be present, um, what liability could uh is going to exist, and so on. But the way the student starts is they have to interview their client. They have to talk to the client. And uh so there are a series of facts, uh, and there's no pre-written dialogue throughout the entire platform. Um different from, let's say, you know, the the world of Zelda or whatever, any other game, uh Dungeons and Dragons or anything, which has a script. There is no script. There's facts, there's facts of what the deal are, there's facts of a party's positioning, there's facts with regard to what the adversary knows that you don't know. All those facts and law um are documented on a deal-by-deal basis of what you want to put in the module. And then we have uh these various personas. So there's the client, there's the tax advisor, there's the adversary, there's the adversary's tax advisor, and then there's the there's the adversary's client. And this can be expanded into multiple other personas. You might have an environmental issue, so you're gonna have an environmental attorney on one side, an environmental attorney on the other, and then you may have an environmental advisor, somebody that's doing the phase one and phase two reports. You can build out any scenarios that you want. And uh there are a series of law firms now that are using this um uh to train uh their their, I'll call them, you know, young associates. Um, and the purpose is is that um you learn more by doing, um, but this is a risk-free doing. Um we all had well and all of us are rather senior in in our and in our tenure with regard to the practice of law, and we were sometimes thrown in uh to the hot water and negotiated things on our own, but only after watching others, and maybe maybe we were out five or six years before we ever got the chance. This allows uh students much younger, um, or or uh the students of law, young associates, to uh to learn uh by doing as opposed to just by reading. And uh there's some cognitive sciences behind this um that we've worked with uh Dr. Ma with. Um and it's it's actually fascinating when you see it work. Um it took a long time with the early version of Chat GPT that we were on, uh like three uh or two or three when we started, that required a significant amount of effort for us to to break through and have it do the right thing. Um as the models got better, um uh it it was a little bit easier to build the modules and so on. Um, but it's fascinating to watch. And uh and you can actually set the tempo, you can set the the snarkiness of your adversary and so on, and it can it can be a lot of fun.
SPEAKER_01Uh well I have to I have to interrupt. If you can set the snarkiness of your adversary, is there also a dial you can turn to um uh determine just how many curse words the adversary will uh blurt out during a negotiation? Well, you you just have to use grok because if you use grok instead of you'll see that's right. Oh yeah, you you're right. That would be the way to get that under a larger aspect.
SPEAKER_02Yeah, um uh and actually uh Conrad, um not not this past year, whenever we saw you at the conference, but the the year prior at the conference, um somebody set the temperature you know really hot, and uh and the the the the uh the machine kept walking away from the deal. I'm I'm walking away. We're like, well, okay, go ahead and walk away.
SPEAKER_01No, I'm really walking away. It was like it was it's very good. The other thing you have to make sure you said is the investment bankers in an MA deal say at about 7 p.m., well, we're taking the client to dinner. We'll see you tomorrow. Keep working, lawyers.
SPEAKER_00Um, Charlie, you know, Charlie, there's a there's a tension. You can see it in every one of these conferences, you go to them. There's a tension about what's going to happen to the labor model. I mean, that the the reality is that AI can do, you know, people like Lenny and I, when we were young lawyers, we we sat in on the phone calls, we sat in on on the meetings and didn't do much, but followed and learned and and and worked through the documents. A lot of that work now can really be replaced or automated. That doesn't mean that law firms don't need a junior class. Young lawyers, you know, law law firms are living organisms. If they don't have a junior class, they don't have a future. Don't be like that movie with uh, you know, with the all the babies, there's people stopped having babies in the world ended. You know, it's it's uh so you know what what Deal Mentor does is it replaces that that uh you know that that learning experience that young lawyers had by giving them you know the experience through simulation in a in a in a no-risk environment. And uh and then you're gonna ask me, you know, well, how do you fit that? I mean, the the the the the issue with big laws, that big law is not going to be destroyed by AI, but the model is, because the the they're they're the model of of marking up an experienced junior labor, leveraging junior labor is is is not gonna be acceptable to clients to the next generation because it's it just it it doesn't cost that much. They're gonna have to shift to more value value-based billing like we're doing and we've always done. Um but uh you know, but but you know, but but the fascinating thing is they have to find a role for you know for those junior lawyers and stop looking at stop looking at them as profit centers and look at them as trainees, like other professions do, and you know, get them ready for their turn at bet. And that's what you know, mentor does, and obviously our philosophy with young labor as well.
SPEAKER_02And one last point on that, Charlie. Yep, it's just it's just AI is actually part of the solution as as as as much as it is a partner. So we're embracing the use of AI to help bring along the younger associates because they might not be doing everything um uh themselves. They may not may not have those opportunities because uh a lawyer with wisdom and talent can constructure the deal, uh, you know, draft the deal, close the deal without the help that they used to have. And so therefore they're not they're not touching these things. Now they can go and practice, and the partners that uh that are guiding them can throw more and more and more at them. But that now now that partner is leveraged building the module. Now a hundred associates can now use that module and learn over and over and over again and be challenged. And they can continually build new models, uh new modules and train that younger team. And initially, I we as I said, it was a negotiation simulator, but now uh it's uh it we're still trying to clean some stuff up, but it'll be done probably by the fall. Um, it will be, you know, any subject matter, how to collaborate on getting a consensus for a board meeting, or how to deliver a new product to market uh outside the sphere of the practice of law, but how to deliver a new product to market. How do you build the team to do that? And and who do you have to ask questions of? So it requires communication skills, it requires cognitive analysis, information gathering, and so on. And you can build those uh personas and uh those so that it's not always a negotiation, but originally started in negotiation and now we're branching out.
SPEAKER_01And just slap on a pair of Apple Vision Pro headset classes, call it what you will, and you're in a uh you're in a conference room, if people still go to conference rooms, and uh having uh uh you know a working uh dinner with uh your counterparty counsel. Uh yeah, so I, you know, it's this is very stimulating and and uh and very um much a matter of looking uh uh to the new model uh that within 10 years' time, I'm being conservative, will uh I'm convinced uh revolutionize uh the way uh people charge for sophisticated uh legal services uh delivery.
The Monoculture Trap In Legal AI
SPEAKER_01Let's uh change uh uh our topics a little bit. Let's talk about uh a subject that the three of us have discussed on a previous call. And that is one of the pitfalls that I guess causes you guys and would have caused me had I still been practicing to to gnash our teeth, and that is when the AI, when the technology uh is is not handled properly and starts spewing out things that I think I can't remember Conrad Lenny or both of you said, you know, it sort of regresses to the mean and causes no uh no fewer than than scores of problems in getting a deal done. Talk talk to that, if you will.
SPEAKER_02All right. I'll go first. There's an article that we were publishing, it's called the monoculture, or otherwise known as homogenization trap of the in the practice of law. Um and it's gonna be published by uh uh by Thompson Reuters, and there'll be a version of it on our website at flatironlaw.ai. But um monoculture, the the phrase comes from agriculture and other places. Uh, if you plant the same type of plant in one location, there's a higher likelihood that it will fail because pathogens will then wipe out the entire thing. Uh uh, I read an article that took that idea and talked about it in the first instance in the financial services sector with everybody using AI to write training programs. So they're vibecoding or using Claude Cowork and building training plat trading platforms. And those trading platforms are trading in the market, and it's it's has a great concentration risk issue because it's it's basically those systems are trained on the same set of data, and it's going to make the market significantly more volatile. In the practice of law, most of the uh of the models are trained on in the big sense of the word, they're trained on a big set of legal data, maybe whether it's agreements, regulations, or otherwise, and they it tends to uh produce results that are the mean, not what your client needs, but just the mean. Mark up this document for me, it'll mark it up, but anything that's in your in either your version or your adversary's version that's not in the mean um is going to be a red line. What happens in the monoculture trap is that you don't realize that you're all you're getting is the mean. And if you have a client, you don't want the mean for certain things. Maybe you'll accept the standard language on X, but you have leverage in in some way, or you have something that you have to get. So you do not want to essentially live in the mediocrity of what the models will always offer.
SPEAKER_00Okay, you want something different, you want something better, you want the reality of uh uh I mean I mean Lenny, I mean there may be there may be areas where you can live with the with the means, NDAs or software licenses. But but for the high value work, it you know, you you you hire lawyers to give you a better result.
SPEAKER_01Right, tailored a bespoke tailored result, tailored to your your leverage, your needs, your risk profile.
SPEAKER_00Sure. And and the other the other big risk factor, I mean it's a bunch of risk factors, but another big risk factor is that is that you know, unless you unless you very carefully prompt the model, the model's not going to appreciate you know your your your leverage position. It's not gonna appreciate nuance, it's not gonna appreciate, you know, judgment. You know, if you ask it, you if you ask it to give you comments on a document, it'll it'll give you uh it'll comment on every single section and it'll go to what you know with what it views as the most rational result. And we've had very, you know, real live instances of that. And we've heard about it, you know, a lot of other firms as well, where clients will then take a document late in the deal and and and run it through Claude and come up with a bunch of comments, which, you know, maybe intellectually would be great, but the pro the client has no absolutely no leverage to it to achieve that result. And if we were to actually submit that markup for to the other side, would it put a lot of stress on the deal? And and uh it's a it's a common issue now because clients think they're so smart and they'll they'll run things through Claude or whatever they're running it through. And we have to tell them you don't have the leverage.
SPEAKER_01Yeah, and and you know the models intellectually may have a perfectly uh beautiful approach to uh how that markup and red line should go. But of course, they don't have any context about the deal.
SPEAKER_02Uh and and and on when on that point, Charlie, it it everybody talks about hallucinations. It's not the hallucination issue. And I just want to make that clear. This is not a hallucination. It the point is that it finds um it finds this mean, um, and it may, in the abstract, sound great to the model, but it has nothing to do with, as Conrad just described, has nothing to do with what's rational. Uh, and that rationality comes from judgment of time, understanding deal making, understanding the transaction, understanding the event, whatever it is, and recognizing the reality. And this is true across legal research for litigation or or otherwise. Uh the other a related issue to this is that oftentimes when there is a repository of information and you ask uh any one of the models, doesn't matter which one, give me an answer, it will give you the best answer. But the best answer ordinarily is not the comprehensive answer, it'll give you the best cases, it'll give, but it'll miss this other case that would have been meaningful. It will give you the best clauses in these documents, but the other clause that it just omitted because it wasn't in its statistics the best answer. So comprehensiveness matters more sometimes than best. Leave the best to be determined with judgment, leave the make sure that the model is uh accessing and evaluating comprehensively, which they naturally don't do. You actually have to force them to do that.
SPEAKER_01Well, that's the key. I mean, as you said, it's a good distinction to make. It's not a hallucination, so that that's not what we're talking about. But as you said just now, it's it's really uh not necessarily uh uh that it's wrong. It's just not appropriate. And that sense of appropriateness can only come from uh the uh the the human being with the wet computer between the other.
SPEAKER_00And all that said, Charlie, it it also makes mistakes. I mean Claude made a mistake this week for you know on a transaction that we were working on, you know, and and I wound up having to go to Delaware Council because I didn't trust the answer that I was getting. Um so that is also you know it's a true risk. You know, now all that said, Charlie, the the efficiency potential of AI is transformational. You know, Lenny and I closed a small MA deal just a couple of minutes before we got on this call, and it was so small we didn't get and we didn't involve any other labor. And here these two 35, you know, plus out, you know, uh uh senior lawyers, we ran everything. That's closing checklists, you know, the uh the the initial drafts. Obviously, we contributed to we we we reviewed everything, but we were able to essentially close a deal, just the two of us. And uh and and there's a you know a dramatic efficiency that uh we couldn't have done even five years ago.
SPEAKER_01Absolutely. Absolutely. Well, uh gentlemen, it's it's uh I'm jealous of my listeners' time, so we we're gonna we're gonna uh bring this uh invigorating discussion to an
Where To Find Flatiron And Closing
SPEAKER_01end. Um if people want to learn more about each of you and uh and what Flatiron's doing, uh what's the best way for them to find out how you think and uh and uh what your practice is all about?
SPEAKER_02Yeah, you uh everybody can go to Flatiron, which is just like the building in New York, flatironlaw.ai. That's our website. You can read about us there. We're also uh on LinkedIn, uh Conrad Everhard uh and Leonard Noir. But Conrad publishes more on LinkedIn. I'm usually sitting in the back room coding. So uh you'll see a lot of things.
SPEAKER_00You can go to go to our events page on the website. You'll see all of our podcasts, all of our articles. We you know, we we we we've written in Thomson Reuters several times. If you want, you know, more thinking, it's all there, you know.
SPEAKER_02Well, or pick or pick up the phone. I mean, we like the old-fashioned way. We like talking about the phone.
SPEAKER_00Well, he likes talking on the phone. Um you know, it's that's that's an antiquated model.
SPEAKER_01I have mixed feelings about that myself, Conrad. But uh I I certainly spent a lot of time on the phone. Well, guys, um I I can't say when you get to New York, I'd love to have you look me up, because you're in New York by and large. But we'll be getting together um, I'm sure, soon at an event. And uh once again, thank you for I can only best describe it by saying stimulating, invigorating discussion. Uh, I know our listeners are going to enjoy it. And um stay well, both of you. Well, thank you, Charlie, for having us.
SPEAKER_02We really appreciate it.
SPEAKER_01Thank you for listening to the Legal Tech Startup Focus Podcast. If you're interested in legal tech startups and enjoyed this podcast, please consider joining the free Legal Tech Startup Focus community by going to www.legaltech startup focused.com and signing up. Again, thanks.