A Moat Big Enough For A Boat

In this episode of Further Comments, we discuss the latest developments in legal tech, AI, and the challenges and opportunities they present. From the impact of Claude's new legal plugin and the concept of vibe coding, to the evolving roles of lawyers and coders. We also explore the implications for access to justice and the need for reliable and useful AI tools in legal practice. 

We also promised the listeners the link to Anna Guo's studies. Here it is!
https://www.legalbenchmarks.ai/home

00:00 Current Events
01:06 Rule of Law
02:43 Claude
04:16 Moat and Competitive Edge
06:33 Workflows
14:02 Accuracy
22:16 Market Impact and Future of Legal Tech
24:52 Customer Needs vs. Wants
25:54 Innovative Legal Tools: From Queries to AI Agents
27:47 The Innovator's Dilemma in Legal Tech
30:42 Vibe Coding: Revolutionizing Legal Tech
35:27 The Future of Coding and Legal Practice
42:09 The Role of Managers in the Age of AI

Transcript

0:01

Hello everyone and welcome to Further Comments. Thrilled to have you here, and I'm always thrilled to be able to talk to my friend, Horace Wu. Horace, how are you? Damien, it is wonderful to see you. I'm doing very well. It's been a while. Like, every time we do these now, it's like "Hey we are supposed to do this every two weeks!" And and then it's like five weeks later we go "Did we forget?" We, we did forget, and there's, there's also been a holiday in there and also something small. Greetings from the Twin Cities where,3000 federal troops have been in my hometown, and my wife has been kept up by helicopters, at night. So, so there's also that. We're not gonna talk much about that, but worth noting that there are stressors all over the world and, my part of the world, there's a bit of stressors too. Like sometimes legal tech just doesn't, doesn't feel like a big deal when there are these sorts of things happening in the world. When the rule of law which is a fundamental building block of society is being — well, I'm gonna use this phrase— being absolutely shat on at the moment. Um it's is that is that a saying in the US or is that just in Australianism? No, I think that's a Latin term I think. I don't remember if it comes from the Latin or the Greek.

There's a real question as to: 1:06

should we have things like adherence with the First Amendment, adherence with the Second Amendment, adherence to the Fourth and Fifth Amendments. These are basic building blocks of our democracy. And I think that we should hold everyone to account, of these things. We, we hold these truths to be self-evident. I think that adherence to the First, Second, Fourth, Fifth and other amendments should be self-evident. And I'm, I'm hopeful that our people and our governments adhere to those rule of law principles.'Cause if, if we don't, then, number one, our democracy is in trouble. And number two, what good is legal tech if AI, we have robots that can be able to tell you the law that won't be followed. Yeah What value is there to a lawyer... to tell you the law that won't be followed. We should really worry, whatever end of the spectrum you are, worry about whether you've curried enough favor with the executive to be able to get what you want. Rather than the executive and everyone being held to the rule of law. Whether you're on this side of the aisle or that side of the aisle. The rules are all we have left. So I, I'm hopeful... I and you signed this letter in support of the rule of law. I, I hope that we can in a bipartisan way — and we had people from all sides of the political spectrum sign that letter— this is an important aspect of our democracy and our microcosm of legal tech. I cannot agree more. None of this matters. Nothing matters If there's no rule of law. We might as well go back to the days where we carry around clubs and live in caves because essentially the same; but we are not gonna dwell on that. We will not. This this is not why people tune into us. Um although you know if you if you tune into us to talk about politics. We we we are happy to do do that all day. Um but we are gonna talk about what's happening in in law and in technology and in AI, cause it's been a big month, Damien. It, it has, it's been a whole year, fit into a month. These are exciting times and especially with Claude releasing cowork and then the plugin called legal. They've said the quiet part out loud. So what, what do you think about Claude releasing the plugin called legal? I think Bob Ambrogi's calling this uh the Claude-pocalypse. This affects legal tech more than it affects lawyers. And it also affects consumers a lot more than it affects enterprise. So if we do that kind of like quadrant, where you got X-Y and A-B, for enterprise practice practicing lawyers I don't think it's gonna change their lives all that much. For consumers looking for legal help I think it's gonna be a massive step change. Because now they can get a lot more legal information, not necessarily legal advice, but legal information directly out of Claude. And people are gonna start building MCP connections, API connections into Claude so that they can pull from the right sources. So, the Midpage guys are already doing that. They've already released an MCP connection into Claude. And that's gonna continue progressing. On the legal tech side, people who are doing thin wrappers, I think your days are numbered, if not up completely. And the people building sophisticated tech, they're gonna be extraordinarily happy that this has happened because it's gonna allow them to go a lot lot further and spread the technology to a brand new class of consumers that previously may not have had access. So that's a very quick summary of my views, but Damien what's your view cause I'm I'm sure you have thought about this long and hard as well. I have indeed. I've been thinking really since I joined Fastcase back in July of 2019, and then, which turned into vLex in 2023, and now it's turned into Clio most recently. I've thought about from 2019 to the present, what is my moat as a lawyer, and what is my employer's moat as an employer. Because I only wanna work for companies that have a decent moat that will keep me gainfully employed. So I think Claude plus legal has really exposed the moat. Our moat at Clio, and before that at vLex has been the cases, the statutes and the regulations. We have yesterday's case, yesterday's statute, yesterday's regulation, and we tell you that yesterday's case overruled last year's case. So having recency and also, you know, authority, that has value that Claude plus Legal does not have. That moat became all the more clear. And then I'm also grateful that I also work for a company that has businesses of law data. How profitable was this most recent case? What's our most profitable areas of law? Maybe we should chase those more. That is something also that very few other companies have, and even the big two whose names, you know, and the other big, big-ish two that everyone talks about, they don't have that business of law data either. So I think that you're right that the wrappers should be quite worried because they have to say, okay, what makes me different than Claude plus Legal? I can be able to crank out a bunch of litigation workflows, a bunch of transactional workflows, a bunch of M&A workflows. But so can Claude Code, and so can Claude CoWork. They can be able to crank these things out too with their ingestion of the entire internet plus a bunch of cases. What value do you have as a legal tech product? So I think this is a bit of a reckoning. Kudos to Midpage for, for jumping on this. You'd mentioned their MCPs. You could be able to take a"if you can't beat 'em, then join them" kind of approach. I think that they've certainly done that. The third thing, the last thing before I kick back to you is we've talked on this pod about the importance of the stuff that's stuck in lawyers' brains. That is stuff that Claude and Legal does not have. As within your organization, within your law firm, within your legal tech companies' former lawyers that have experience in funds, or M&A, or litigation. To take the stuff stuck in those lawyers' brains and then be able to create workflows that wouldn't be known in Claude plus Legal. I think that stuff stuck in the brain is a moat, at least for a little while. So I think those are ways that any legal tech company and any law firm can kind of compete against Claude plus Legal. But, but what do, what do you think? Oh so many thoughts. So let's let's start with the the lawyer's expertise and these workflows. I, I think we've seen Legora and Harvey both trying to deploy these portals; these client facing interfaces. And part of why this potentially can work is if lawyers actually dump their expertise into pre-made, prefabricated workflows that their clients can access and and make use of. It's not gonna get to the same level as an expert lawyer but that's not the point. The point is it's self-service It helps people get a certain part of the way there for a much lower fee. We'll talk about fees in a little bit cause I've got lots of ideas on that. One of the dangers here is, of course, well if Harvey and Legora capture all this, what's the value of the lawyer then? Can't they sell direct to in-house teams. And, and this is one of those thoughts that has been discussed a lot among both law firms and in-house teams. And I don't think we, there's a landing on that. I think it it it comes down to like, you know, what's relative value that lawyers bring that these AI workflows can't bring; but there is that potential risk, right? So that's that's number one. Number two is you have Claude and GPT being essentially the means of production that all these legal tech companies that are using generative AI... that they have to leverage. So what is stopping Claude and GPT from saying "you know what, we are gonna release GPT-6 internally or Claude 5.0 internally, but no one external will have the access to that"? Then they control the means of production and lock out the market. Is that gonna happen? Maybe. Maybe not. So, lots of uncertainty about that. And then there's this idea of like, well, how does that shape, like, "what lawyers can do to make themselves relevant?" And we've talked about integrity and trust and there's a third point that uh Jordan Furlong... and integrity and intuition is the third. You can't have GPT in your ear while you're arguing in front of court or around the boardroom. You know tho those are those are all kind of ideas. Um and in terms of moat... uh I had a really interesting coffee this morning with Nick Fleisher from Sandstone. So he's he's one of the uh startups that just recently got funded 10 million bucks from, I think it was uh someone big who was it... Sequoia! We sat down for a coffee We exchanged some information... he was telling me how what they're building effectively becomes a command center for in-house legal teams. So they're not simply giving people a chat bot that interfaces directly with the language model. They're connecting everything up using the language model as a means of thinking and funneling and channeling and triaging, so all the all the hard engineering that has to happen to make the data flow and the work flow forward. And that was a really clever way of I think using the technology to create a moat because I don't think Claude and GPT are gonna come out and create a command center. And using agents in the raw is not gonna be reliable enough. So I think that's a really clever way of like engineering a solution out this. And finally my fourth thought and you know I'm gonna stop because otherwise I'm just gonna rant all day on the Midpage thing,"If you can't beat them, join them". Legalweek, in a few weeks from now, there's a day that Nikki Shaver, Chris Ford, et cetera are organizing for vendors. So rather than just selling to others, the vendors get together talk about what they do and et cetera. And Oz Benanram has organized a panel where Otto from Midpage and I are actually gonna be speaking on that panel about vendor collaboration. And the panel is called"Build, Buy, or Bond". It's this really cool idea of like, well, if vendors get together, if vendors allow others to tap into their capabilities and their data, et cetera, what does that future look like? So, Otto for Midpage clearly is selling the data. We at Syntheia selling our analytical and comparison technology as kind of a baseline that other people can leverage. So... lots and lots and lots of things happening right now in this market mainly because it's been shaken up. Everyone's trying to find their place in the world right now. If GPT can do everything, what's left? That is the right question to ask. And you'd talked about means of production. You think back to Marx that, you know, the means of production historically has been land. Whoever has the land, has the agriculture. And then that migrated from land to humans, in the early 20th century, now that is the means of production. What we're shifting now, what if the means of production shifts from humans to whoever has the GPUs and the electricity: the intelligence. If the means of production moves from the human intelligence to the machine intelligence, whoever has the GPUs and energy — so that is the Microsofts, the Nvidias, et cetera— they control the means of production. And the value moves from the human side to just a few players that have that means of production. I worry about us losing a lot of human jobs, pitchforks coming out, and then we have some societal unrest. So that's, that's thing number one. Thing number two, more close to our legal tech home, just this week met with a, a big company whose name, you know, that I'm, I'm not gonna say, that has said, let's take all of our litigation tasks, every single thing from pre-suit investigation, all the way through appeal, and let's just automate it in-house, so we could do all this work in-house and then send the draft complaint maybe to external counsel and, say, "tell me how this is wrong." Or not. And get it out the door. Or maybe in pre-suit, we just decide what the liability is or lack of liability, and then settle for a certain number, before we even call a law firm. So anyway, they're looking to do this work in-house. And you talked about a command center, and that kind of thing for law departments. I guess another anecdote is that, you know, Salesforce said on stage with me at ILTACON last year, they said in the first six months of last year, they already saved $3 million in external spend that they brought in house. And they were anticipating between $5 and$6 million for the entirety of the year. So that's $5 or $6 million that a law firm just is bringing in house. So I think that what Sandstone is taking advantage of I wonder if the migration is a huge amount of legal spend on law firms, that keep raising their rates every single year, without fail over the last 30 years. I wonder if that's reached now a tipping point that the corporations say, you know, we can now have this amazing intelligence that is just gonna bring the spend from external to internal. And the law firms they just won't hear the phone ringing because the, the in-house counsel won't need to call them anymore. I wonder if many firms are realizing this and realize that this is a bit existential for them. To be able to say "no, i, as a law firm have stuff in my brain that you as in-house counsel don't. And I'm gonna be using the tools that you also have access to. But I'm gonna be able to then augment and offload my brain into these tools to be able to do this work." We talked about "you can't beat 'em, join them". Uh, I think that law firms and their clients are probably gonna be using similar tools. And the law firms really have to leverage their knowledge: "this is what I can do, that you as in-house can't." And then the real question is how much delta is there for that brain trust in the law firm? If I as an in-house counsel can already draft the complaint — with all the cases, statutes and regulations that are relevant — what is the value then of my law firm brain on top of that? Because in the past, I used to make money on drafting that complaint. But if I don't do that anymore, what is the value I provide? Nobody knows the answer right now. Everyone's trying to work out what the right model is. I, I had this revelation. I, I wrote about it, but we haven't spoken about this. I, I was talking to an innovation person in a law firm and we were saying how five years ago, before generative AI, when you tried to sell software to law firms and you said "my piece of software is amazing. I can get you 80% accuracy!" The law firm's gonna close the door gently and say I will never see you again. And, and yet right now, anyone who's selling Gen AI, we all know behind closed doors, 90% is like the best you're gonna get. There's always gonna be that little margin of error. And yet every firm is jumping onto it. So we're like this is interesting. This is weird. What, what happened there? And we kind of pried apart and pulled all the layers off the onion. And the the observations kind of follow this path of: okay so there is a 20% sort of like error gap or potential error gap. Everybody knows that. Are clients happy to accept that 20% error gap? The answer's no! Not a chance! They're not paying law firms to give em crappy advice! So that this gap has to be addressed by law firms. Are law firms doing anything other than throwing humans at the problem and the answer's no. They're still throwing humans at the problem and that doesn't scale. Because the more work that an AI can do, it just means humans have to review more and that time is limited. So how do we close that gap? How do we either let humans do more you know or what do we selectively do with human time that allows people to charge more money and still sustain themselves, et cetera? And where we kind of landed, and I won't go into my entire blog post, but where we kind of landed was I think firms are gonna have to start stratifying the work that they do. So they will do some of the really high-end work. And they'll say for this work our billable hours are gonna be higher. And then they're gonna do some really low end commoditized work. And they'll say, Hey but you know what, this low end commoditized work we're gonna charge you less money to do that work, because you can get that elsewhere for lower cost as well. And I think within the same law firm, even one partner or one associate will start having differentiated rates based on the work they do. So this was a very interesting I think potential development. Have you have you seen that, Damien? Is that happening out there? Yes, I am seeing that. And also this reminds me as it so often does of Jae Um's and Ed Sohn's "Cream, Core, Commodity". Because what you've described is essentially: "We'll take the Commodity stuff and we won't charge as much," as you just said, for that Commodity things. And the real question is how much of "we won't charge you as much" goes up to the Core?"We love to have 'bet the company' Cream stuff, but really we make most of our money as a law firm with the Core." AI has already taken the commodity things. And it's inching up to the Core things. And the real question is how far up into Core and then into the Cream is AI going to give us? Those are all unknowns. Another thing I wanna talk about is your point about accuracy. The, you know, 90% accuracy— or maybe 80% accuracy. How does one define that accuracy? That is, how do you write evaluations, right? Because you know 90% of what? And you know, what is the 10% error? Because do we have a Platonic ideal of the "best contract"? Or a Platonic ideal of the "best brief"? And you ask 10 lawyers, is this a good contract or a good brief? And you'll get 15 different answers. You know, when we talk about"accuracy" in data science, we talk about missing things. And in data science, we call that lack of recall. Or did you get things wrong? In data science, we call that precision. So really the question is, can you be able to identify precision and recall when there is no Platonic ideal of the"best brief" or the "best contract"? At best you could be able to say, well, that was an error in the law. You know, they, they said that Roe v Wade was good law. Okay? That is an error. Therefore, okay, now I can put a 90% number on 10% of getting the law wrong, that is a 90% accuracy. And if you missed a bunch of cases, that is 10% recall. But I would say that, a lot of these generative AI things, they're not getting the law wrong. They're not saying that Roe v Wade was good law and they're not, doing, something in a contract that is gonna scuttle the deal. What it is is saying, I as a lawyer, think this thing sucks. That is not objective, that is subjective. And another word for that is called taste. Rick Rubin, the legendary record producer, he said, I'm not a good musician. What I have is good taste. People come to me because I could be able to say, "Hey, I. I like this and I don't like that." So I think that's what we as lawyers do. We're like, oh, that contract sucks. Well, why does it suck? Oh, because that was written horribly. Or that that was ambiguity, right?

Well, but then another lawyer: 18:32

that could be the best contract on the planet. So I, I think that we really need to be careful when we say, you know, 80% accuracy and 90% accuracy. Is that really identifying that 10% of it sucks according to my taste? Or is it 10% is in error, that is bad with the law. The winners going forward are going to cater to the user's taste.

To say the law: 18:54

Roe v. Wade is bad law. It was overruled by Dobbs. And I'm gonna tickle your brain in the way your brain likes to be tickled, and give you the "taste" that you like. I would ask you like, how does one measure 80% accuracy or 90% accuracy? Uh, or is it just taste? So, you bring up a really really interesting point. And it's one of those funny things where, en masse, if you were to test like the correctness of the output of generative AI, it may or may not fluctuate between those numbers, right? And correctness like you said is hard to measure. However, the better measure that I think people should be moving onto is usefulness. How useful is the output for the task at hand? I think this is a callback to Jack Shepherd who wrote about this a long time ago. Which is I don't care if it's 1% accurate if it's 100% useful, It's awesome. And I think there's this kind of like old world thinking of where you have machine learning classifiers, where a piece of data is relevant not relevant, pertinent not pertinent, et cetera. Like very easy to measure accuracy in in those cases. But with the generative AI outputs where it is far more flexible, far more subjective, it's hard to draw a bright line and say is it accurate or not accurate, unless you are doing one of those almost binary tasks where you can measure correctness. But I think that's really hard with Gen AI, or at least with the sophisticated gen AI use cases, accuracy's not really a measure you have in the first place. That's right. And your talking about usefulness reminds me of Anna Guo, who is, I think, out of Singapore. She's created an evaluation scheme that we'll link in the show notes.

The evaluation scheme: 20:36

the X Axis is reliability, and the Y Axis is what you just talked about, usefulness. So what they did is had human lawyers draft legal documents, and I think they were transactional documents and litigation documents, if I'm not mistaken. And then they had humans do it and then they had ChatGPT do it, and Gemini and CoPilot and Claude. And for those who are watching on YouTube, you could see that human lawyers were both less reliable and less useful than ChatGPT — and far less reliable and far less useful than Gemini out of the box. So this is a double blind study where human lawyers did the thing, Gemini did the thing, you mix up the results, gave them to human evaluators, who said that already Gemini out of the box was more useful and more reliable than humans. Um, so I, I. Why Why is Claude all the way at the bottom when it comes to usefulness? Why? So this is September of '25. If you were to put this in Claude Code or Claude CoWork, I think you would have a far different answer. Because they would have Claude's agents be able to do this work. So September of '25... there is a timeliness or un timeliness of this. This goes, what you're seeing on your screen here, clients, want "good enough." They don't need perfect. They need "good enough." And really right now, according to the human evaluators in the study, Gemini was not just "good enough," but is actually*better* than the human lawyers. This goes back to our conversation just a few minutes ago, talking about: where does Claude Legal fit into this? And is Claude Legal "good enough"? For that, mom and pop, company? Or for that Fortune 500 or maybe Fortune 1000 company that doesn't have a lot of money for lawyers? Or for that consumer that just got kicked out of their apartment? Or that person that's worried about losing their kids. What is "good enough"? Every lawyer needs to be thinking about "good enough". So context for those who are listening to this like a little while later a little while after we release it. In the first week of February, Claude released a set of skills that can be used by Claude Cowork, and part of those skills were legal skills, things like reviewing an NDA or marking up things. And it sent shockwaves in the legal tech market. The shares of major legal tech companies like Thomson Reuters, Lexis Nexis, LegalZoom, et cetera, drop between somewhere between 15% to 20% each. My personal view is it's an overreaction in terms of share price drop, but underneath I think there is a hint of validity to the fear. Which is what's the role that these companies are gonna play if Claude becomes so pervasive and useful? Also surprising that the market didn't already price this in since we saw this telegraphed from like a year ago, but here we are. The Claude-pocalypse is people questioning what's left for legal tech if Claude out of the box can do so much. That is the right question to be asking. And whether that market drop was founded or unfounded, I guess time will tell. A correlative data point is that you may have heard that our friend Pablo Arredondo, left Thomson Reuters. Yes, Uh, I jokingly said to uh someone, said, "oh well, it was a two and a half year earn out, not a two year earn out!" It's true. So anyway, so these, we'll call them legacy legal tech companies need to step up, if they can, and show what value that they do have that essentially would validate your claim that 20% drop was unfounded. And if they don't step up, and query whether, you know, with that kind of market cap, like what are the funds to do R&D, or are they looking to tighten their belts and maybe shrink a head count rather than adding head count? I think it's on the incumbents... it's incumbent on the incumbents to be able to demonstrate their value. And if they don't come up with new and innovative value to provide atop the Claude Legals, then maybe the market shrink will be validated. I'm gonna present a mental challenge for both of us to talk about, and we we didn't raise this before the call. Given how quickly people's expectations are changing, given how quickly technology is moving, what's more important? Addressing the consistent long-term pains that customers have always expressed, or building and shipping technology fast enough to keep on the frontier? So, in other words, where are we focused? Are we focused on the evergreen problems or are we focused on the changing tastes and preferences and needs that keep coming from customers week on week? That goes back to Henry Ford — and maybe it's apocryphal, but Henry Ford apparently said that " if I'd asked customers, they would tell me ' I want a faster horse.'" And Steve Jobs, similarly, if he'd asked customers, he wouldn't have, created the iPod, which are 10,000 albums in your pocket. So this is the idea of what the customers want isn't necessarily what the customers need. Imagine a large company saying we would love for a tool that we have internally as an in-house counsel, I would love to be able to collect the corpus of the emails, and the Slack messages, and the documents, and put all the relevant documents into this corpus. And then let the in-house counsel run queries across that corpus to be able to say, based on the claims, breach of contract, and the elements of the claims, that there was a contract that was breach that you performed. Right? All of these things. I, I as a user want to be able to run queries across this, to be able to fill in each of those buckets. That's what I think we should do. So that is what the customers expressed that they should do with this tool to be able to query this corpus. I said, yeah, but we live in a different world right now. How about you have the agents, that is Vincent, go through and say, what are the causes of action in New York and California? What are the elements of the claim in New York and what are the elements of the claim in California? And then take the facts from the corpus of documents and put them into those elements of the claims, into those buckets? And do that programmatically! And then after you've done that, say, gosh, in California's bucket three is empty. And then create queries, questions for the business people to say, Hey, do you have anything to fill up bucket number three, 'cause we need bucket number three. And then have the AI do that for you. And then serve that on a silver platter to the lawyers. So option one is have the lawyers run queries across the corpus. Option two is the agents just do that programmatically and serve it up on a silver platter? Option two is obviously where we're headed and where we are right now. But option one is what the customer wanted. So they want that faster horse, but we really need to be thinking differently. And so I would say that, good product people, and I would consider you a great product person and I, I consider myself a product person. We need to be able to say, let's go not just what the customers express, not what the customers express as their need, but what is the underlying need. Not the drill, but what is the hole in the wall? Or more importantly, how do I get the photo on the wall, and not even have to worry about, the hole. So really I think we can now have robots not just put the hole in the wall, but build the entire damn house. There is the, the end and there are the means. Customers will often express means, without expressing the end that they want to justify those means. Really our job as product people is to say, what is the end product? What is the, what is the output that we're looking for? And give them that output, regardless of the means expressed, but give them the output that they need. So the reason why I asked this question is because we are talking to law firm innovation folks, in-house teams, and one of the consistent themes that come up is, at least the last two years before the Claude-pocalypse killed

wrapper startups, is Innovation teams: 28:02

Do we bet on the startup who's able to continuously pivot very quickly continuously listening to feedback? Or, do we bet on an incumbent who has been building towards the same ob objective for the last decade, right? That's a consistent question that people ask. You you are, you are essentially saying, if a product manager has the right vision, is listening, is building towards that, whether they started from the incumbent or from a startup, it doesn't matter. I think that's right. And the question is, do those innovative product managers, do they live more often in the incumbents? Or they live more often in the startups? This is part of the innovator's dilemma, right? Once, uh, you go from startup to being now an incumbent, the startup person may have cashed out and leaves. Did we mention the Pablo left Thomson Reuters? Um, anyway, so. dunno. Did we? Right, right. So, so then the real question is for the incumbent then, what kind of innovation is left? And are there product managers with incentives to be able to think differently in the way that we just described? So the question is how many of those innovators have product managers that are gonna push the edge? And are they incentivized to do that edge pushing? Or are they just going to collect their paychecks? And be able to collect their monopoly rents or duopoly rents? This is why the Innovator's Dilemma by Clay Christensen, and if people aren't familiar, they should look it up, this is how it happens. Then you have a startup with fire in their belly Yeah To be able to say, we are able to do things faster, better, stronger. Go ahead and stop your contract with the incumbent and come with us, because the incumbent has lost the fire in the belly. Well, I mean that's that's what Sandstone and Chamelio are doing in the CLM market and disrupting all the incumbents who can't build and pivot fast enough. Yeah. Uh, that's right. You know, if you're a duopoly, why would you invest in R&D? Why, why wouldn't you just collect monopoly rents? This is, business-wise, rational for the incumbents to not invest in R&D and just continue, reaping monopoly rents. And you can imagine that these are multinational corporations that don't

just have one business line: 30:15

legal is not their only business line. They also do tax and accounting. And it turns out that accountants and tax, they pay money for software. Where lawyers hate paying money for software, because lawyers bill by the hour and if I spend less time on an efficiency tool, I make less money. But tax and accountants, man, they do flat fees. I go to an accountant with my taxes. They say it's x number of dollars, uh, annually, and if they shrink their time, they make more money. Well, speaking of disruption, everyone now can vibe code and I think what that's doing is it's letting law firms, like we we don't exactly know where it's gonna go, but it's potentially opening up the path of: One, you can definitely experiment a lot faster. You can validate ideas a lot faster, and prototype. Two, potentially these become the backbones of internal products that otherwise would've never seen the, the light of day. And, we don't know where it's gonna end up because we don't know how competent and how secure these things are. But if Claude gets better, if all these models get better I think maybe that will become an endpoint. For those who've been paying attention Jamie Tso is a lawyer at Clifford Chance who's been vibe coding a lot of these. And he's essentially started a movement, and it's it's exciting to watch. Damien, uh what's your reaction to the vibe coders? I, I, I love, love, love this. And for those, who are watching... if I'm faced this way, this is my work computer. If I'm faced this way, I'm playing music. And something new since the last time you and I talked, Horace, is if I look this way, I have my vibe coding setup. Where I have a lit, I, I do, it's a third computer that my wife is very, very sad that we now have... I have 1, 2, 3, 4, 5 screens, 6, 7, 8, 9, 10 screens, uh, in our third floor. So, uh, 10. You're like in the Matrix. It, it is, you know, I, I'm plugged in, I'm jacked in, and so on this side, these two screens here, have a Linux machine, with Claude Code installed on it. And I'm just cranking out stuff left and right. I'm cranking out FOLIO things on the nights and weekends, and all the things that we've been talking about that I've been waiting for Bommarito to build, I'm just building it myself. And, I am maybe one 100th of the coder that Mike Bommarito is, and one 100th is even being generous for me. I'm probably one in 1000th of the coder he is. But I have taste. And I know the requirements for what all of my SALI and FOLIO people have been asking, "Hey, maybe you could do this, maybe you could do that." And literally, I have a microphone set up over here where I just speak into existence. I said I need to create a tool that could be able to map an ingested taxonomy from a law firm, and then be able to map it to the FOLIO outputs. And I could be able to say what I want to do first, is for the large language model to be able to identify what maybe looks like a location versus a document type versus a jurisdiction, for a court, et cetera. So once you do that, then you could be able to then put tags around that. To be able to say, this looks like a jurisdiction, or this looks like a document type, or this looks like a forum. And then once it does that, then do a mapping to FOLIO items, connect those to the Python scripts and the other queries to be able to map that document's motion to dismiss with FOLIO's motion to dismiss. Or the document's merger agreement with FOLIO's merger agreement. I literally speak just like I'm speaking to you right now, and all that gets transcribed. And then that turns into a product requirement document. Much like I would build, as a product requirement document, for all of my employers. And then in the past, I would have to throw that PRD, that product requirement document over to the coders, and then they would build, but it turns out, Claude Code is happy to take that product requirement document and build. Part of the vibe coding too is that I say, Hey, Claude Code, based on what I've just spoken to you over the last 15 minutes, now ask me questions that will give me a better products requirement document. And they'll ask me questions like, do you want it to be local to the machine or do you want it to reach out into the internet? Or a local model like DeepSeek or Kimi do you want them to run locally on the MacBook? So it'll ask questions like that and dozens of others. And then I'll say, keep asking me these questions and we'll keep refining together until you don't have any more questions. And then once there's no more questions, then it just cooks. And then we have a product. So I, I think that people that haven't tried vibe coding, in the way that I just described, really don't see not just the future, but the present. They don't see what Jamie Tso and others are seeing. What do you think, have you, have you been vibe coding? I vibe code to the extent of getting a design in the front end which then my engineers take off my hands and say please don't touch it ever again. Because we, you know, we we already have a a massive code base. And I don't want to just inject something that doesn't fit with what the developers need, cause then maintenance becomes impossible. So, a little bit different. We're not starting from scratch here, we will create to the extent that we can, get all the right things in place, and then the engineers would put it into our style. So much like a law firm, you know, whatever clause you come up with let's put into house style. Let's let's, you know, reference the right definitions, et cetera. So that's where I get to with vibe coding. The few times where I've tried to get a a fully functional working application, we got it working and the engineers will say: don't. Stop. Just just stop. We, we cannot use that. Not only are there lots of functional issues, there's a lot of security issues as well. We have to be a little bit cautious just because we can't plug things in. But if I was starting a project from scratch the world's your oyster right now. Like going into any of these tools; You've got ready-made tools like Manus, where you can literally just dictate and it'll put the entire thing together for you. Or a little bit more low level, you go into Claude Code, and you can control everything a little bit more. It's super duper easy as long as you have a crystal clear idea of the problem you are trying to solve and the user experience that you want people to have in solving that problem. That's exactly right and we, all we turn from heads down coders, of those of us who are heads down coders to architects. To be able to say, what is the problem to be solved? And then architects always consult with the product people, who are closest to the problem to be solved. And so I think what this migration is gonna happen is to go from coder to architect to product people. And really even the hardcore coders will just turn into product people. People may have heard that Claude Cowork, which is Claude Code for everybody else. Anthropic who makes Claude, they built Claude Cowork in a week and a half. And they used Claude Code to be able to build Claude Cowork. And so, Not a single line by humans apparently. Not a single line. They, as product people, say this is what I want it to do, Claude Code, go ahead and code it up. And even my friends in the Valley and others on podcasts, some of the best coders in the world say, I'm a really good coder, and it makes me really sad that I'm never gonna write a line of code ever again. This is something that, uh, and, and I, I think there's an analogy here for lawyers. Because coders are craft people. They are very proud of this beautiful code that they've done. So in this way that they're artisans in the same way that lawyers are artisans. I'm so proud of this contract. I'm so proud of this, this motion. I as a lawyer, am a failed novelist, and the way that I express my art is through the motion or through the contract, right? So I think that a lot of lawyers and coders are both artisans, and what the coders are seeing is now a canary in the coal mine for the lawyers. Because the coders are saying, gosh, I'm sad that I'm never gonna write another line of code. But I'm also excited about all the new code that I'm gonna be able to build. So this, this kind of mourning the old world, but being excited about the new world. I wonder if lawyers are gonna mourn the old world. To say, gosh, I'm never gonna write from scratch another contract, and I'm never gonna write from scratch another motion. But think about how much more legal work I'm gonna be able to do. Now let's go to Legal Tech and software as a service.

Old World, you had software as a service: 37:54

you built once and then you go to 10,000 people, say with Salesforce, right? So you build one thing to be able to go one to many. But with vibe coding, it's one-to-one. Yeah, this is essentially software for me. I don't need to worry about serving 10,000 people. I don't need to necessarily worry about the security problems. I know exactly the environment I'm in and it's right over here in my laptop, so I don't have to worry about what, does it work on Linux or does it work on Mac, or does it work in Microsoft? I have software for one. And so what does that mean for SaaS? For when I can do software for one? And can I just create throwaway apps? And as legal tech, even established legal company, can I just do throwaway code? That I shouldn't be so precious about all of the legacy things that I've spent years building up? I shouldn't be precious about that because code is cheap. It's, I can do a million, tokens for 50 cents. And, and that's tens of thousands of lines of code for just pennies. Maybe what the future of coding is for legal tech companies and maybe for law firms is to... you know, you talked about house style: this is the way that we as a company, or we as a firm, this is our coding house style. Maybe that just becomes context. Where you throw Claude Code at it, saying use this house style. And here's a PR, a pull request, and be able to use the house style to be able to do the pull request. And I think that our coders and our lawyers are both gonna have to get rid of that artisanal mentality. And not be so precious with the code that we've spent decades building up and the legal writing that we spent decades building up. And just go with the vibes, man. Because that's where the future is. And not just the future, but the present, like Jamie Tso demonstrates. Two thoughts on that. One, I read on LinkedIn, I cannot remember who wrote this, but there was a it was a GC and she wrote, Using Claude, using GPT, she can get something that is like 90% good enough and you can just plug it straight into a document. And the only thing holding her back is she'll just stylistically go well that's not how I would write it. Or or I might have to tinker with that definition just to make sure it lines up. But in terms of the effectiveness of what it drafts, it's right there, and... and that goes to taste. That goes to taste, right. So if the tools are already good enough but it just doesn't fit with the style, unless there's some market practice that mandates you can have a consistent style, consistent drafting as you do for ISDA master agreements et cetera, or ISDA schedules, why don't you just plug in what the model's draft out for you, if it works! If it works. So that's thought number one. And and I think as you say, lawyers have to get over this hump of, Hey I do the best drafting in the world. If it works, it works. Full stop. Second idea. And, and by the way, I can already feel the burn in my inbox from all lawyers coming to me saying what are you talking about: If it works, it works. There's a better way of doing this. Guy, I hear you, I know. But second point, second point is, so we we do consulting with law firms sometimes, where we say to them look here's what we think of the universe, how your stack can evolve, where it fits in with knowledge and AI, and blah blah blah blah. One of the things we've observed is, over the last 12 months, it used to be: we wanna train your lawyers on prompting, on getting the most out of these models. That's almost completely gone now because the models are so good and so robust in interpreting even bad prompts. And also there's functions now you can go Hey improve that prompt self auto prompting. So prompting's gone out the window but what's becoming more important is now we want lawyers to train, train on managerial skills. You wanna manage your lang, language model in the same way you would give instructions to a junior, or the agent in the way that you can manage it to do task one, tell it when to stop to come back to you, so that you can funnel, review and then send it to agent number two, and so on. And that sort of skill of management is becoming much more important. And then the follow on question is when's that gonna change? When can AI self-manage and what comes next? So I think that this this is evolving so so quickly. I love that. And yeah, All of the coders that I've been talking to say yes, I've spun up seven, eight different instances of Claude Code. And I'm managing, just going from instance to instance, managing, saying, uh, you've now come back to me, now do this. And then I go to the next one, you've come back to me now, do this. And I think that what we're gonna see, both with the coders and with the lawyers is the good lawyers who are good managers. That is the partners that have given specific directions to their juniors. And you know, not have the junior saying, gosh, I have no idea what that partner wanted. Right? Those are, those are bad managers. The good managers will be concise. And they will be clear, and be able to tell their juniors — whether those juniors are humans or whether those juniors are bots — to be able to do this work. Managers in law have never been really valued, because if the juniors spin up a lot of time and waste a lot of time and bill a lot of hours, that's, that's not a lot of disincentive. Oh, that's too bad that they spun up and wasted a lot of hours. I guess I'll bill a client for as many of those hours as I can, right? I think that the good managers for humans are gonna also be good managers for AIs. So that's thing number one. Thing number two is, you talked about prompt engineering is not a thing anymore. And I, I wholeheartedly agree. It reminded me of, you know, taste we talked about earlier., And for those who are looking on YouTube and those who aren't looking on YouTube, you should look on YouTube. Where this is my default, this is part of my system, prompt. Uh, saying avoid passive voice. Write clear and concise sentences. Eliminate unnecessary words. Do not use "be" verbs like am, is, or were. Favor active verbs that are strong and specific. If you're gonna use a date, start with that date, on June 10th, this thing happened. Avoid preposition phrases. Favor "parties' arguments" over "arguments of the party." If you think about how much that person that you talked about saying it was 90% there, it just, I needed, it didn't write like me. If you say. Here's how to write like me. And by the way, this comes from Strunk and White, and this comes from Bryan Garner. This is not writing like Damien Riehl. This is just good writing according to Strunk and White, and Bryan Garner, and Flowers, and to all these other people. People will soon realize that their taste is really universal. By the way, all the things I just showed you are baked into Vincent, which is why Vincent is such a good writer. It is a much better writer than ChatGPT out of the box or Claude out of the box because it has these good writing tips in them. As we go forward, I think the idea of taste maybe is going to come into question. That maybe, "my good taste" and "your good taste," there's universality to that. There's so much more we can talk about, but we are almost at the end of time, Damien. And I'm going to ask you the question that you usually ask me, which is what's what are you optimistic about? Man, I am so optimistic about just vibe coding in general. I'm optimistic about our customers doing things on their own. And then that spurring us, as legal tech companies, to do things even faster, better, and stronger. You know, Ford Motor Company, Darth Vaughn, said, Hey, law firms, you have a new competitor in town. You know who it is? It's me at Ford Motor Company. Because you have to say what you're doing with the AI more than I can do with AI. So I think in the same way, I'm excited about my customers, law firms and law departments, vibe coding all over the place. Which then says to me, what can I do above what they can do? And will spur me as a legal tech company to do things faster and better and stronger and all of my teammates to do things faster and better and stronger. So I'm excited about vibe coding and being able to crank things out more quickly because our customers are demanding it. What? What are you excited about? I'm excited about Claude and GPT opening up the access to justice market for everyday consumers. Where they'll be able to just by saying, Hey I need legal help, Claude will automatically switch to a legal persona, tap into the right data sources, tap into the right tools and say how can I help you? I think that is on the horizon, if not already here. And that just helps so many people that historically has never had the help. that couldn't afford the help. I think we are we are now at this moment. Now what does this mean for law firms? I think it means very little to top tier law firms. But it means a lot to everyone who's now providing service to consumers. Where you have to step up. You have to do more. You have to do it faster, better. Create and provide that differentiation and it's gonna drive everyone to improve. It will be hard, but we'll all end up in a better place. I hundred percent agree. And future that you've just described works only if the cases, the statutes, and the regulations are free. That's the only way that this works. So here's hoping that the governments, the courts and the legislatures and the regulators, put all of those out in the public domain. If so, we will reach the future that you and I both hope for. I, I love our conversations so much, Horace. Uh, thank you so much for, for having them. And thanks to our audience for joining us. This is great fun. Uh yes, And let's do this again very soon Uh Legal Week is coming right up and the next episode we're gonna do something, something like a retrospective episode. I'll go and pull together the pieces, and then let's see how much of what we said six months ago still holds today. I'm sure 100%. Uh, thanks so much man. So good talking to you. Thank you everyone for listening. Bye. Thanks. Bye.