The Things, They Are A-Changin

LegalWeek Wrap: GenAI Valuations, Agentic Workflows, Build vs Buy, and the Future of Legal Work

Damien and Horace record an end-of-season-two LegalWeek conversation about how AI has shifted from flashy banners to an integrated reality, alongside soaring valuations for legal GenAI platforms Harvey ($11B) and Legora ($5.5B), which they contrast with the legal-content arms of Thomson Reuters, LexisNexis, and Wolters Kluwer. They discuss agentic workflows (including runaway agent costs), a Fortune company’s in-house buildout of over 160 agents, and the ongoing build-versus-buy dilemma. They map the legal AI stack (data, foundation models, UI/UX), note transactional data’s immaturity versus litigation, and anticipate consolidation among 1,000–4,000 legal tech companies. They explore adoption patterns at top firms, portability and lock-in, workflow/IP questions, niche tools displacing broad platforms, and concerns about a reported 90% drop in graduate offers in Australia.

00:00 LegalWeek Reunion
00:37 AI Everywhere Now
01:22 Unicorn Valuations
02:41 Agents Run Wild
04:05 Build Versus Buy
05:07 Legal Tech Stack
06:32 Too Many Startups
08:10 Workflow Patents
10:26 Valuation Math
12:15 Chasing Services TAM
14:08 Law Firm Hedging
16:23 Data Portability
17:25 Graduate Apocalypse
19:46 Model Ceiling Debate
22:51 People Process Gap
24:21 Free Versus Services
26:32 Open Source Over Vendors
27:22 Time Value Versus Laziness
28:42 Utopian Deflation Future
30:48 Legal AI Products Maturing
32:35 Jevons Paradox Legal Market
33:19 Consolidation And Moats
36:32 Paying For Consulting
38:40 Marketing Stunts At Legalweek
40:29 Copyright And Workflow Theft
45:02 No Moats In UI Cloning
47:41 Build Fast Get Noticed
49:27 Niche Tools Displace Platforms
51:10 Infinite Forks Open Source
52:59 Season Wrap And Cheers

Transcript

0:01

Damien. Horace Wu. So good to see you in person. Hey, I think this is, uh, end of season two. Mm. Very exciting. And then we were just talking before we started recording. It was, it was two years ago at LegalWeek, and here we are at, uh, in the middle of LegalWeek, two years ago where we said, before even the podcast, we said, Hey, let's, we have great conversations. Let's record each other's, share it with the world. No, that's right. And, and turns out, here we are two years later. And it, it, it seems like five or six or 10. It's time is a circle. Oh man. The last couple of years have just like sped along beyond anyone's expectations. That's true, but the time is accelerating. And, and you know, like we've both been at LegalWeek this week. Yeah. Um, I've haven't had as much, I, I haven't had as many conversations as normal. Um, and I think partly it's because I'm tired, and partly I think the market is tired. Yeah. I walked the floors today and I didn't see as many AI kind of banners as I did last year, but it's all been integrated. It's all just now folded. It's, it's the air we breathe. There's the joke that's, uh, said, uh, the fish said it to the other fish, Hey, hey, how's the water today, boys? And the other one said, what's water? Right? That's brilliant. Right? I mean, it's, it's everywhere. Yeah. Yeah. Well, here we are. Here we are. Uh, and you can't get away from it, and it seems like there's no end in sight. And, uh, as of yesterday, our, our, our favorite legal tech, Gen AI platforms are combined worth $16.5 billion. Favorite? Everyone's favorite. Everyone's favorite heart throbs. They're on Teen Beat Magazine, AI Legal Teen Beat Magazine. Oh, if they were only right? So what's the number? How many billion dollars? Well, precisely, I think it's a squiggly line, $16.55 billion. 16.55. But, but who's counting? But that's a, that's unicorn times 11. Well, that's right. Wait, I'm not sure that's times 11... but that's, it's remarkable. Really. It's crazy. I, I scratch my head. Um, many are behind closed doors. A lot of people are like, how do you justify that number? What's the TAM? Where does it all come from? Smarter people than me have figured it out, I'm sure. Have they? Yeah. Well maybe. Maybe everyone's scratching. They're putting real money behind it. Yeah, I, I guess that's true. It's almost like Polymarket where people are, people are putting real money, but is real. Is it, is it real or is it crypto? Like what kind of money's coming back? Well, is crypto not real? Is that what you're telling me? Are birds real? So yeah, the whole market has kind of gone crazy and topsy-turvy. Um, I heard some staggering numbers. And I, I was sharing this with you earlier today. So earlier today we caught up. We, um, uh, saw Rob Saccone, who's yes now. You know, out there creating stuff. Vibe, coding like crazy. He's got different agents going and he, we were talking about how, uh, how we were fear like a fomo, like everything we're doing we're like, what's my agent doing is? Is my agent asking me three questions that he should answer to be able to do the next thing? It's wild. Pretty soon, I think it's gonna be like, you know what? You go and figure it out. Don't come back until it's done. Yeah. And if it's substandard when it's done, I'm sorry, I'm firing you, I'm getting a new agent. Maybe. But then, so that's, we'll call that one path. But with me, I like, I, I like to catch it before it goes too, fa, far off the path. Mm-hmm. Like I don't want it to churn for four hours and come back and give me crap. Right? I'd rather admit a 10 moving the right way and then let it keep going. Right? I heard a fantastic story of this startup who like ran all these agents on Claude and on GPT and uh, two of the agents just started conversing with each other. Hmm. And it wasn't monitored and the end of month bill was $47,000. They had a really good chat though. It was so nice. Really great chat. So it's gonna run away. Yeah, yeah, yeah. Oh goodness. So yeah, so Rob Saccone is doing amazing things with vibe coding and uh, I was also, um. I talked today with a, a large fortune company whose name, you know, that I'm not gonna say. Mm-hmm. Mm-hmm. Uh, but they said that, uh, within their law department, uh, legal operations team, they've already built 56 different agents. Wow. Uh, that they do internally in, and they, they, that's already done in the, in the can. Uh, they're doing 50 more currently, and they have 57 more in the pipeline. Wow. So between the three, that's over about 160 different applications all the way from rifle shot, tiny applications to more broad spectrum applications. Uh, so if this is indicative of people all have a build versus buy question, at least this law department, corporate law department, is saying, we're gonna build as much as we can. Uh, and then the real question is, do they need to hire? Do they need to buy our two favorite? Companies that are worth $11 billion maybe. Yep, yep. Uh, or do they just need to build themselves? And what value do those two favorite companies have above what that in-house team is doing? That they know exactly what they need because their lawyers are the ones that are doing the vibe coding and solving their lawyers' niche problem in a way that a SaaS company can't solve that lawyer's niche problem. So I, I was breaking down the kind of market, the technology stack, if you like, and at the very bottom you've got the data layer, right? Mm-hmm. You know, whether that's, uh, case law, we talk about this all the time, whether it's internal data of your contracts. And then you've got your intelligence layer. The LLM. Um, and no one's competing on that, that that's Claude, that is GPT, right? No one will compete on that. It's gonna be super hard to catch up with those guys now. Nobody has the billions. Yeah. Uh, and, and then you've got the UI UX layer, which is kind of where everybody's focusing their energies. Mm-hmm. And competing. Yeah. And essentially, like if you look at our, you know, favorite platforms plus all the other tools out there, they're all competing on the UI UX layer. Yeah. There are very, very few who are really competing on a data layer. You know, you've got Clio, who's got vLex. Uh, you've got Thomson Reuters. You've got LexisNexis. You've got Wolters Kluwer. Okay, great. And then, tho, those are the kind of, you know, li litigation side. Mm-hmm. Right? The, the case law. And on the transactional side, you've got DeepJudge. Yep. You've got DraftWise. You've got probably Centari. And then you've got us at Syntheia. Yep. Uh, I can't think of any, anyone else. Yeah. And by the way, like on the transactional side, we are nowhere near as mature as the litigation side. There's just not as much data. Who's sharing. Right, right, right, right. And then there are a billion legal documents on the litigation side, and a billion is hard to come by on the contract side. If only! Right, right. Uh, and, and. With the number of companies, uh, I just read or heard at, uh, on a Monday, uh, I was at a, a thing, uh, at Cleary. Uh, Monday said, they said LegalTech Hub, the number of legal technology companies is now over a thousand. Uh. I thought it was 4,000, 4,000 legal tech companies? I, I mean. Uh, Nikki can, Nikki can tell us, yeah, 4,000, oh, Nikki, please tell us what number it is. Right? But, but still, either whether 1000 or 4,000, which would be remarkable. Like that is, I re I, I thought it was just a few months ago, was 750 or something. So like, like we we're just as the technology rocket ships up, where we just rocket shipping up with the number of technology legal tech companies. Yeah. Well I, I was meeting with a, a friend or a new friend who's starting his own legal tech company. He does litigation stuff, basically turns messy litigation process into one structured process that can be kind of uniformly applied to all litigation and therefore you can standardize the data, right? So I was like, I've never heard this before. Wait, standardized, like SALI standardized? FOLIO? I asked the same question I asked the. Same question. I will send you his details afterwards. Yeah. Um, and, and he's like, no, no, no. It's not about tagging. Tagging is like one aspect of it. Okay. It's more like mapping to a process. Mapping to, you know, this is kind of like overarching steps 1, 2, 3, 4, 5, 6. And there's 1(a), 1(b), 1(c), 1(d), and there's 1(a)(i), 1(a)(ii), 1(a)(iii). And you can kind like map it down. Yeah. Along this process. Yeah. And that's what he was doing. Um, and I thought it was a really neat idea. Um, but that sort of company, which I thought was like super duper narrow and niche, that exists. Uh, what you've just described. Yeah. Uh, there's a patent, uh, where I'm an inventor for Thomson Reuters. Okay. That describes that very same thing. I think this company should know about that. And they might want to know about that. And because, uh, the patent, and I'm not speaking outta school, but the patent is that you take any, any, uh, litigation or transactional matter. Yep. Break it up into a thousand parts. Right. And then of course, uh, that thousand parts has a subpart as a sub subpart and a sub sub subpart, and just map that all out and then be able to say, this is the way you do this task. This is the person down the hall that has done this task. Here are the queries that you, in your lead legal database to do those things. Uh, that that is all in a patent, uh, filed in 2015, uh, with, uh, so still valid 'cause it's 20 years, right? So, so there, there's a real question as to, you know, how many of these things, right? These are, these are, difficult times to be able to think about. Uh, well, the real question is, is that patent valid or is it obvious? Because how many years before 2015 have we been doing process mapping? Right, right. And how many times before 2015 did you apply that to a computer process? Yep. Um, and so, uh. I would say that the thing that I'm the inventor on, there's a very good argument that one could very well make, but there's a, don't argue against yourself here. No, no. There's just, you know, maybe hundreds or dozens of prior RS that maybe invalidates that particular pattern that that's up to the patent attorneys to go and fight that. That really is, in the meantime, you are the inventor. We can say that. We can say that, but, but really the real question is like the workflow is the thing. Yep. And I, I, I've been thinking for at least 10 years that, um, you have the task, but then you could have a subtask and then sub sub subtask. And, and what is a task is really in the eye of the beholder because you could all go all the way down to, uh, you know, the secretary called the courthouse to

be able to see if they're open at 6: 9:53

00 PM Right, right, right. One task is, you have to pick up the pen. Yeah, yeah, yeah, yeah. So it's turtles all the way down. You haven't heard that phrase for a long, long time. Yeah, yeah, yeah, yeah. That's true. It's true. So anyway, this, this is kind of like, you know, the landscape now is just busy, busy, busy. Yeah. Um, and, and like last year we talked about Wes, we talked about how tired everyone is. I think this year it continues to be tiring. Yeah. And I think there is now this push towards greater adoption. Mm-hmm. More training. We'll come back to that in a second. Yes. Yeah. I actually forgot to finish my point from earlier. Okay. Which was, um, $16.5 billion. So crazy. Um, a combined value for Harvey and Lara. And, uh, someone told me, uh, that the combined value of Thomson Reuters, Lexis Nexis, and Wolters Kluwers's legal services and legal content arms combined is $4, $4.5 billion. Four and a half versus 16 point some, uh, and uh, turns out that Thomson Reuters LexisNexis has some content... and every lawyer in the world is a subscriber. Yeah. And so you kind of go, well, hang on a second. How does the math work? That math ain't math. There's additional compounding information, which we talked about too, right? Which is, if you look at the fees charged by TR, mm-hmm. Westlaw for a subscription per year versus a Harvey per year. For some firms, one of them is four to five times more the than the other, and it's not the one you think. Thomson Reuters is the one that's four or five times more expensive than Harvey. Bizarre, but, but the valuations flip flop. So weird. Anyway. Smarter people than me. Well, but, but okay. So what, what's the tam then that, that 11 point or 16 point of whatever? Billion, well, I guess 11 for one, and then four. What, how does it work out? Uh, uh, so Legora is now value at 5.5. Yeah. And Harvey's valued at 11 billion. Okay. So, so 11. So, but really between. You know, TR's, you know, whatever that valuation is versus 11, where's that delta? Uh, so is that going direct to consumer? I mean, where does that money come from? Between if you're not selling, like have we exhausted the, uh, lawyer market? Well, one argument, and I, I steal this argument from Brad, uh, [Blickstein]. Yeah. And, and the argument is it has to be services. They, they can't be going after the TAM of software. They have to be going after the TAM of all legal services. Right. Which is, I think 400 billion. Right. So, and, and some of that might be eating the launch of lawyers. Yep. Uh, but some of that might be tapping into 92% of unmet needs, uh, because we lawyers are too expensive. Yep. Yep. And uh, and the question is, uh, do lawyers go down market into that 92% or the, the Harvey's and the others go down market? Oh, you, you're thinking they are Okay. Yeah. Well, I mean, there, but it's down market. It's a new market. Right, right. It's, it's a, yeah. It, it is exploiting, uh, in the best way that market. Okay. I don't know. I don't know. I, I just, I just feel like I'm missed out by not investing on day one. You were asked, I'm sure. Hello? Do you have $5?$10? Yeah. And like, but I, I wonder if, you know, we've been wringing the efficiencies out of every other area of business, uh, and then the investors are like, oh, law's expensive. Mm. Uh, let's invest in that. And maybe the, you know, the math ain't mathing and doesn't matter to the people saying, No, we'll figure it out in the back end. Let's, let's wring the efficiencies out. WeWork kind of like philosophy, right? Get the market first. Yeah. Yeah. Yeah. Who knows? Maybe. So weird. Who knows? In the meantime though, I think, I think like the legal market and the legal profession's benefiting, um, I mean when there's a lot of money put being poured into any industry mm-hmm it benefits. Yeah, it's true. To what extent do you think. Uh, so walking around, of course, the legal tech companies are, are, um, frothy with excitement and money. Uh, but, uh, and, uh, law firms, uh, seem a little more, uh, a little more edgy, I found. Yeah, yeah. Are you finding that? Uh, I think, so, you know, we work the AmLaw 100s, the AmLaw 200s. Yeah? And, and here's an observation we are making. So there's a small bracket at the top. Let's call it 10 to 20% who have bought Harvey, who have bought Litera. But at the same time, they're also buying subscriptions to Claude and OpenAI with the intention that, hey, if the foundation models, if the GPTs, et cetera, catch up, then we switch over. Yeah, right? I'm sure this is not a surprise to you, Max or Winston. Um, and, and if Harvey and Legora managed to keep on staying ahead of the models, then the law firms stay with Sure. The, the Gen AI platforms, right. That's 20%. And I think they're hedging their bets. They're keeping themselves open, they're training the people, getting the data ready, all that sort of prep stuff, right? Small, small, minority. 80% are buying Harvey and Legora and spending their entire effort rolling it out, ensuring his adoption. And I don't think, I'm sorry, I don't think they have a long-term vision of what they should be doing. Maybe I'm wrong. Uh, uh, hard to know that, that I think that an, anybody. Has a difficulty seeing the long-term vision of what is going to happen three months, six months, nine months down the line. Well mean, look at last month. Right? Right. Claude CoWork came out and everyone just Right. And, and, uh, the month before that OpenClaw, uh, and then, and a few months before that Opus 4.6, which just turned the world around to be able to say, oh, you don't need to be a coder to be able to just crank out production ready code. Yep. So, uh, so I, I think hedging bets seems to be the logical way for a law firm to be able to say, nobody, nobody sees three months in advance, so let's just buy lots of things and Right. And, and hope that, uh, one of these we'll be able to port our current workflows from system one to system two to system three. Right, right. And like a quarter million dollar investment today by law firm to have an enterprise license to Claude for everybody to use is a really, really small premium to pay for that kind of insurance. Yeah. Yeah. And, and I wonder a lot about, uh, GDPR. Requires data portability. Mm-hmm. Saying that I, I, I should be required to, uh, be able to take as a consumer from, uh, social media site number one and be able to port that data into social media site number two, I wonder a lot about GDPR requiring the same of our two favorite companies and others. To what extent would GDPR for our European friends say, Hey, I'm the European that wants to be able to port from system one of our friends into Claude? That only benefits the European though, doesn't it? Well, maybe unless you have the lowest common denominator, like a lot of, a lot of companies will say, well do it for everybody because we have to do it for the GDPR anyway. But I would guess that maybe they'll make an exception in this case. Well, there's a lot of money to be made by not making it portable. Right? Right. And one could imagine maybe it's their demands to, to from consumers to be able to say no. If we're gonna subscribe to you, it has to be portable. All questions to be answered. So much fun. But I think in the meantime, everyone just kind of trying to figure it out. I'm hearing, and I haven't shared this with you, uh, so hearsay. Take it with a pinch of salt. The best kind of say. So, uh, a friend of mine who is, uh, a lawyer in Australia. Uh, he or she tells me that in one of the major capital seeds in Australia, the graduate offers that have gone out this year to lawyers has gone down by a certain percentage. Now I'm gonna ask you to have a guess. What percentage do you think it's gone down by? Between last year and this year? So. Uh, so I would say the conservative me wants to say maybe 7 or 8%, but the realistic me is gonna say 15%. 90%, nine zero. So the, the graduation apocalypse is happening. It's happening. Uh, the singularity is here. It's happening. So like, let's say a firm sent 50 offers out last year to graduates, five went out this year. That. So all the law students that are saying, am I gonna have jobs? And I'm saying, Jevons paradox, maybe I'm totally wrong. Or maybe I was trying to lull them into a lullaby. Like, you just put your kids to sleep. Right. You can tell your 2-year-old, oh, you, you're gonna have a job. It's gonna be fine. Yeah. Twinkle, twinkle little star. You'll have a job where the sun sets tomorrow. Right, right. Oh, so I, I just am one of my wife's friends. She's just graduated from law school. I hope she's not watching right now, but she's, um, my wife is very worried about my friend's future. Yeah. As am I. Yeah. I think it's my friends. Uh, yeah. I mean, she spent a significant amount of money, like over $100 or maybe $200,000. Right. For what? Starbucks? McDonald's? No, that, I mean that, yeah. It's what happens. Honestly, I have a view that this might be like COVID. When COVID first happened, people didn't have jobs for those two years. Right? Or, you know, the, the graduate classes for those two years really thin out and then people recognize, oh crap, now we don't have any third year lawyers. And they went on a hiring spree. If they need third year lawyers instead of their workflows that they've been coding up and boring. Well, that question is, where do we hit the ceiling for large language models, right? Like at the moment we still hit ceilings. We see these in more complex workflows. Yeah. We see these in longer documents. We think there's like three factors simultaneously come into play to determine whether or not we cross the ceilings or exceed the ceilings. One is, can we keep on increasing the context window? Yeah. So more and more information can go into the model at the time. Yeah. And I think that will happen. Yeah. Every 18 months or Moore's law, right. I'm not sure this is six months or 18 months. It doubles. Yeah. Increase the context window or do more context to engineering to fit more into that. Right. Compression. Compression. So, so either way. Yeah. More, more data into the window. Therefore you can add the more documents, become more complex stuff. Yep. Second is, can you increase the accuracy, get rid of the whole loss in the middle effect, and so on. That's is scientifically not as certain as increasing the context. Google's trying. Google's trying, Google's trying hasn't happened yet though. Um, and the third is, will it be economically cheap enough to be worth it? Yeah. Right. Like, like would it make sense to throw in 10 documents to go, Hey, model, go figure out what's market for me, versus chunking them first, throwing in just the relevant parts and then saying, figure out what's market from these relevant parts. Yeah, yeah. These, these are all true. And then I guess, yeah, I wonder, are we going to, as you know, people, we've been saying, well, this, uh, 200,000, uh, token window, then expanded to a million with the Google. Right? And that was within a year. So we thought, of course, in a few years we're gonna have 2 million tokens and 10 million tokens and right. 10. Uh, but we've, we've, uh. I was just listening to Dario Amodei, who's uh, I think on the Dwarkesh podcast, and he was saying that, um, if you, uh, to be able to do that, to go to two to 10, et cetera, um, that is gonna increase the duration of your training cycle.'cause you have to train at 10 million to be able to do inference at 10 million. Yes. Yes. Uh, and so, uh, and who's gonna wait? To train the 10 million tokens, uh, context window, uh, because then it's gonna take so much longer that you're gonna fall behind the Groks, and the OpenAIs, and others. So, um, so that's, you have to do it in parallel. You have to train the shorter models and the longer models. Right, right. I thought the problem was there wasn't enough data that have a contiguous 10 million token. That too, right? The, yeah. You have, you know, of course, Harry Potter and others, but is that 10 million tokens? Right? Right, right, right. Yeah. Like, I mean, you can synthetically create it, but then we all know the problems is synthetic data. Yeah, yeah. So, and then, then, uh, these are all pre-training rather than post-training, which is the RL, uh, re uh, reinforcement learning. So, yeah, I, I don't have, we reached an asymptote, uh, asymptotic? Maybe, maybe, maybe, maybe we, we have, but, but I think you were saying this before, and I agree with this. Like even if the models stop advancing right now, right? There's still five years of development. For sure. People, process, technology. Even if the technology was stopped, the people in process are gonna need to catch up. Yeah. Which, you know, as, as we rocket ship up, you know, I, I think that we're, uh, really gonna see, uh. The people in process f fall farther behind. I don't know if I, if I agree because I, I, I was looking around the floor today and I think the people part I can't speak to, but the process part, I think that's what a lot of vendors now building towards, they're integrating process into their tools. Mm-hmm. And then using the large language model intelligence to drive these steps forward. Sure. That's, that's, well, I guess. There's the technological pro process, but then there's the, within my law firm, I've always done it this way for the last 25 years. I always give it to Jenny, who then gives it to James, who then does the thing who are now all agents. Right, right. Well, may but Jenny and James need a job. And are we gonna fire Jenny and James? Well... maybe. Right? Right. But, but I mean, I like, no, I always put the stamp on the thing, and that's what I do. And boy, if we don't do that, the hell freezes over and, and, and then bad things happen, right? So, so I think that these kind of human processes, uh, it's hard for the vendor to be able to work those things out. I agree. I agree, although I think, I think there are some vendors who are trying very hard to capture those human processes into their workflows. Yeah. So, uh, yeah. It it's up in the air. I, I, I wonder, I, I was talking with a, uh, large law firm, uh, he's not a lawyer, but he's part of innovation mm-hmm within, within the law firm. And he was, uh. He said, I said, what are you seeing that's interesting on the floor? He said, uh, there's a company, I'm not gonna say the name because I, I don't want to cast dispersions on this. But one of the 4,000 companies out there, what they're doing is, uh, doing what I did on a weekend to vibe code, to be able to say, upload any text, whether it's litigation or transactional, and put all the tags on there and structure the data. So this, this non-lawyer, uh, innovation person was saying, and it's amazing, then you can have really well structured data and be able to then use that to fuel the large language models. And then I broke out my laptop and said, I mean like this thing that I did last weekend, that's all open source, that I made open source and free. And I said, how is this company that you're so enamored with. Going to compete with free, because I put it onto, uh, the, you know, the GitHub and you can download this today. Uh, run it on, it's a Windows app. Unplug from the internet and just do it locally. Like, you know, you know how they can compete? Services, they'll provide the services at a white glove fee, but people will feel good about it. And that, that's exactly what he said. Oh? He said, he said, Damien um, lawyers aren't nerdy enough to do what you just described. And I said, nerdy enough to go to a website and download an .exe and double click and then start doing the thing. Uh, and he said, well, no, but there's liability there because if I download it and do it, and I muck it up, then I'm responsible. Whereas if I buy this service and they muck it up, I say, no, get it right. And in a sense that's what you're saying. Like I get to point the finger at the responsible person who will fix it. And whether you call that a technology or a ser, white glove service, the thing's gonna get done without me having to do the thing. The liability shifts on. Yeah. A little bit. But then I said, but okay, let's play this out. I said, so you, uh, say to this, uh, company, I like it for a lot of things, but it sucks for this, that, and the other thing. Um, but then they say, we're not gonna fix this, that, and the other thing. And then you're stuck. Yep. Option two is to download my thing. And then you say, I like it, but I wish it had this, that, and the other thing. And you literally say to Claude Code, make it have this, that, and the other thing. And because it's open source, all of a sudden you have those three features, 20 minutes later. So between convincing this vendor to do something and just speaking it into existence, why wouldn't be we going this way? I agree with you that there was, there was a possible future where law firms and internal engineering people will say, you know what, we can just vibe out the solutions. But if, if human nature plays a role here, a lot of people would just say, you know what? I would rather pay X dollars to have someone else take care of that for me. And then we're gonna end up in the same boat. That's right there. There's time value. The value of my time. Yep. And the value of mine was more laziness. Well, same idea, right? I mean that like. Uh, there's a, a bunch of Mercedes down in the hotel. Uh oh, yes. Right there. So, and they, they're paying a good number of dollars for laziness. Uh, also the time value of that assumedly, billionaire's, uh, value. Uh, so they want to spend five seconds walking to the Mercedes rather than the 25 seconds to go to the parking garage. So. Uh, laziness. Time value. We can call it either way, I think. I think, uh, when you are a billionaire, it's called time value. When you're not a billionaire, it's called laziness. That's right. The dollars on, on each second there. When you're poor like me, it's just laziness. There's a joke that, uh, not a joke, but there was, uh, somebody, uh. Periodically says, okay, for Bill Gates, is it worth his time to reach down and grab a hundred dollars bill? Uh, and that's the same idea. Yes. And I think someone worked out it's absolutely not. You should just keep walking. Just keep walking and let somebody else have that on all, come on, you know, Bill, you have enough money. Don't, don't think. Let somebody else have it. Right. You know, in, in this, we, we we're filming this in New York City, right? Yeah. Yeah. And this is one of those few cities where you can probably walk outside and accidentally bump into a billionaire. That's true. And then he'll punch you in the face. Well. His bodyguard. His bodyguard. Really? That's true. Look, we, we are, we are living in a world where money seemingly just means less and less. So, uh, uh, the Moonshots Project, uh, moonshots podcast, have we talked about this? We, we have. And I started watching it actually. It, it's quite so they. It's the most optimistic podcast you can think of. And they, they say, you know, we watch all these dystopian, futurist, you know, the Terminator and all the things, uh, but we should have more utopian view. Like there is a good path. And a good path is that we just have goods and services just come down massive deflation and $1 will feel like a thousand dollars and, uh, money maybe doesn't matter as much as it does anymore. And with when $1 can be essentially our cost of living universal basic income and universal high income, uh, maybe is more palatable.'cause we'll just take all the taxes from the Nvidia, the Microsofts, and the Googles, which will be running our economy and just spread that out. If the government makes that so. Uh, that is, uh, that is maybe a scenario where money doesn't mean as much, right? Oh, I, I, I, I wish there were more of these drinks to make me believe that will be it. That's one path. It's one of the million, and this is the one that works. It's, it's like that scene from, uh, from Avengers Endgame, right? Yeah. Doctor. Doctor, what's his name? Strange. Goes. Yeah. It's a one in 1.6 million, whatever chance. Yes. That you're saying there's a chance, saying there's a chance! But hey, who, who knows? I, I do hope that's the future. I, I, I hope so too. But, uh, yeah, that, I think that that would require a government for us to, the government to be able to say, let's redistribute and be able to reap the, I'm not gonna say Ill-gotten gains, but I would say reap the, the gains. Yeah. So, so you are, you're saying there's a government, not only is it gonna be, uh, following the rule of law, but also have an ethical sense, and also we will want to benefit the society at large. Yes. Uh, boy, we can, a fella dream. Can't he? But I mean, that's totally more optimistic topics, right? Um, I, I'm actually really excited about like, the progress that people are making this year. Yeah. Um, 'cause I, I remember last year we were having this conversation in the park and, and I was saying, man, I, I was not impressed with Harvey and Legora, 'cause it's just a wrap up. Um, I, I think my view is shifting a little bit. I, I applaud how much hard work those guys have put in. Yep. Um, and, you know, money can do a lot. Mm-hmm. Uh, and, and they've now created really mature, really polished products that actually benefit lawyers. I'm very, very impressed, but, oh, I didn't, there's always a but. I didn't, I didn't see that coming. There's Oh, really? No. No. But. I thought there's always a but. But I think a large part of that is because the foundation models have gotten so, so much better too. And, and, you know, the, the rising waves, the, the rising tides argument, right? Or the boats are being lifted. And to use another metaphor maybe, maybe the waters have washed away many sins of poor, uh, the choices, but then those are just, uh, the, the blessed waters of the AI are taking care of that. Yes, yes. Mistakes don't matter when it's under the water. It's Atlantics, but it's, uh, it's beautiful. May maybe, yeah. So like, so the, the rocket ship of acceleration and capabilities is floating all of our boat. I think so. I think so. And I think depending on kinda like what your strategy is and where you're positioned in the market, as long as I think you plan for six months or 12 months in advance, who, who can do that properly? Yeah. But as long as you try, I think there's space in the market for a lot more than we anticipated. I, I hope that's true. Um, something I've been thinking about is, is, um, is the legal pie going to expand with Jevons Paradox or contract? I, I, I think the people who get help, the, the amount of legal services being provided that will expand. Yep. I'm not sure the money available to pay for legal services will expand. That's a, that's a reasonable, uh, thesis. Uh, yeah, that. And so if that's true, if the money stays constant, um, will the legal tech pie grow or will the legal tech pie remain constant too? Well, if the money stays constant, then there'll be winners and losers. Yeah. And it's not gonna upport 4,000 people or 4,000 tech companies. Well already, so I don't know if you noticed this, but walking around the floor, there were a lot of biz dev and corp dev people from the larger companies, even the newer companies kind of going, Hey, tell me, what are you building? Yeah. And I kind of caught my ears on on a few of those and I was sitting at a coffee, next to one of those and it was a very interesting to kind of wi, hear the whispers of these conversations because I think for the, for the incumbent players, mm-hmm, they have to spend money to catch up. They're now lagging and they're lagging bad. Yeah. For the new players, they have to spend, spend money to keep ahead of Claude and OpenAI because now those are the competition. Yeah. So everybody needs to spend money and I think whatever if 1000, 4,000, a lot of them are gonna be brought up. Interesting. To what end though? What? So I'm a startup that, uh, I was a lawyer that said, oh, this legal tech is gonna be a thing. I'm gonna be able to take all the processes I did as a lawyer and then be one of the 4,000. Um, if I spin that up over a weekend and then roll it out to the world. What's to keep the company that would've acquired that from just spinning that up in the same weekend? Well, okay. That assumes that the new company's building on top of Gen AI. Mm-hmm. Which I think is a very, very small moat. For sure. Uh, I, I don't think those companies. We'll get bored up. I think they'll just evaporate. Okay. So which, which are gonna get bought up? Uh, I think it's gonna be the sorts of companies that have got a, a, a deeper understanding, more domain expertise. Deeper moats. Mm-hmm. Whatever they happen to be. Right. Whether it's technology. Whether it's expertise, whether it is relationships even, maybe, but, but even relationships. I mean, that's a, you know, you and I know lots of people in legal tech and lots of people in, uh, the buyers. Uh, but relationships only go so far, right? You guys don't like me enough. Everyone loves Horace Wu. That is an accurate statement. Oh, no. Okay, so segue. Segue. I learned a long time ago from watching some of my friends start a company. There's a difference between I like you versus I will pay for this. For sure. Breaking out that wallet and saying Here my hard earned cash. That is very, very hard. Yeah. Um, but I will encourage you till the cows come home. But I'm not gonna give you my cash. That is a great idea. Yes, yes. Do not take my money. Please try that. Not that this happens to me, by the way. Right. I just, you know, giving examples and, and, and really then I, when I did pricing, I, I would go out to the customer and say, how much would you pay for this thing? Yeah. And the, the lawyers would say, oh, I, I'd spend at least $300 a month, or at least $500 a month. Right. But it doesn't bear out. But nothing means anything until you have to actually pay money for it. I wonder if there's a PolyMarket for it. Like I'll give you like put your money where your mouth is. I don't know if there's a model there, but it'd be fun if there were, right? I don't know. I don't know if it if it is, but I mean, look, there, there's some, uh, tech bros that very rich off that model, so Yeah. Yeah, I guess that's right. Who knows? But yeah, consolidation's coming. It is coming. And so you, you talked a little bit about services. Yes. Uh, to be able to say that this is a, you know, this is a white glove service that is kind of disguised in their technology. Um, I wonder, as I think about the, uh, the humanity coming to the fore, you know, the widgets that lawyers make, contracts, motions, briefs, pleadings, AI's just doing all the widgets. Uh, so really all that's left is the humanity by going to the fore. So I wonder if there's an opportunity going forward for, um, option one is for legal tech companies to give the tech, but then there's the people and process charts of that triangle that are not the tech. And I wonder if there's a, uh, consultancy play to be able to bring people into the new, uh, the new world. Maybe it's wrapped around some technology or maybe it's just smart people and saying, I'll be your Sherpa through the AI world. I wanna bring back to bring us back to the previous point, which is I love that idea, but I'm not sure who's paying for that idea. Yeah. Um, the solo small lawyer that doesn't know what to do about anything, they're gonna pay, but how much? How much? And then the biggest offerings of the world already have those sherpas in the form of innovation people. Right, right, right. So then you have mid-market. Which doesn't have innovation people, but want to stay ahead. They wanna stay ahead and they're, they're, uh, have more resources than the solo small. Mm-hmm. So maybe that mid-market, there's a, a good play for Sherpas for mid-market. There's always that question of how do you distribute that? How do you make the, that service known to the buyers? Yeah. Right. Cold emails don't work. No. Uh, LinkedIn marketing doesn't work. No. Uh, go into events. Maybe that works. But then are your buyers at those events? So I think, yeah, maybe there is a demand for this, but how is the service provider gonna reach the people that need that service? It's all about distribution. It's all about distribution. Sales, marketing. Hello. That's how... we, we go back to Harvey and Legora. Well, sorry, that go. That's funny. Nobody knows marketing better than them. Well, they're actually true. They're, I love what they do. I mean they, yeah. Like, point to them, hats off to them for incredible marketing that they do. Speaking of marketing at at Legal Week, did you see the Rage Cage? I did. Hello Definely team! That's so smart. Like really, really, but I don't know. Like, I don't know what the symbolism is there. Like, like, are you so tired of your technology that you're gonna take a sledgehammer to it, but here's some technology that you're gonna love. Right, right, right. Go, go and smash that because you've always wanted to. Right, right. Exactly. I, I don't know, I thought it was very clever as well. Um, but the, the Definely team has been. Has been doing some really clever marketing. So I don't know if you've seen on LinkedIn, but their Chief Revenue Officer, Rhys, has been doing like a GoPro Strapp to his front, like walking across Brooklyn Bridge or riding a helicopter and saying, Hey, look at me just preparing this document while I'm riding a helicopter.'cause I'm on my way to my Hampton house. Or I, I don't know the story. Um, but I thought it was wonderful marketing. Yeah, it's, it's hard to, uh, break through noise. Uh, and so yeah, smashing things, uh, or riding in helicopters is a decent way to, and look at us. We're talking about doing now, right? Right. It's true. It's true. Free marketing. Free marketing. Yeah. Yeah. Sam Flynn. Do you know Sam? Yes. Yes. Good friends. He, he and I were just, uh, walking and talking. Oh, yeah. Both Australians. All the Australians, and we each other, just like all the Minnesotans and all the North Dakotas. Yeah. We, we go to school riding kangaroos. The kangaroos. But yeah, so Sam, I, he's just. Been walking and talking. He just had a walk and talk with me. Mm-hmm. So, I mean, this, you need to do things that are not done before. And so I, I felt like I was on comedians, uh, in cars, drinking coffee. Yes. A little bit. And then, uh, and he's kind of got that, uh, the whole jazz vibe. Not to say, Sam, that you're copying, uh, uh, se Jerry. But Jerry call your, have your lawyers call Sam for sure. I don't think you can copyright an idea like that. You can't. Well, you can't. And that Well, and that, that was another thing. Uh. You can't copyright ideas. And, uh, the idea expression dichotomy says that ideas are uncopyrightable. Only expressions of those ideas are unco, are copyrightable. Um, so Rob Saccone and I were, um, walking, um, after you and I, uh, hang out, hung out with him, and we were talking about how much of these workflows are ideas. Yep. Or are they expressions of ideas? Oh, that, oh! So there's a real question like you can imagine if. Uh, somebody works for legal company, uh, legal tech, company X. They walk out the door, start legal x, company y, legal tech, company y, and then from afresh start new workflows that happen to be different expressions of the same idea. And on the backend, the output is identical. But the expressions of the workflows are different. Right, right. From a copyright perspective, they are, you, these are, uh, different expressions. Therefore not copyright. It's not a copy. See, I'm, I'm thinking of a slightly different application of that problem, which is lawyers. At law firms are now giving away their workflows into platforms and these platforms, some of them may be onselling these to other law firms and clients. Mm-hmm. Are they breaching copyright or are the contracts that these law firms signed sufficient to override the moral rights and the copyrights that the author has in that work? Uh, can I put my copyright hat on? Oh, please. Okay. So, so, uh. One could imagine that, uh, I am law firm X and I submit my beautifully crafted workflow to be able to say, this is exactly the way that this thing should be done. Uh, you are law firm y and the platform gives it to you. And now you use the same copy, the same expression of that beautifully trafficked thing that seems pretty clear that it would be, um, like if you make a copy of the thing that is a copy of the expression, therefore copyright infringement. Okay, so that is, but if you just use it without making a copy of it, or is that even possible? Well, I mean, if, if, uh, is your instantiation of the thing and my instantiation of the thing in AWS is that a two copies or 12,000 copies? Right? So they're like, what is a copy? It's the same zeros and ones. Same zeros and one, right? So anyway, so let's say it's the exact same expression of the workflow, same copy, maybe an argument that there is a copyright infringement. But scenario two is you have this workflow that is beautiful and you have a prompt that says, extract all the uncopyrightable ideas from this. What are the concepts that are required to be able to elicit this workflow? Okay. And then take those ideas, the abstraction of you, abstract the ideas. Yep. And then say, now create a new workflow, a new expression of those uncopyrightable ideas. So extract the uncopyrightable ideas. And nothing that AI produces is copyrightable, by the way, according uncopyrightable turtles all the way down. Yep. Extract the uncopyrightable idea. Reconstitute and rehydrate the workflow, but in a different expression of that. Is that copyright infringement? No. uncopyrightable idea got pulled out of it. So you are saying, so long as the Gen AI players take whatever the prompt is and the process is from the lawyer and say, Dear AI, interpret and rewrite that, the rewritten version of that workflow can never violate copyright. It seems like it would be public domain if you listen to the US Copyright Office Public domain? Well, the US Copyrights office says if machine generated, then no copyright. If machine generated, then no copyright. And no copyright means public domain or no copyright simply means not protected. It's synonymous. Interesting. Like if, if there's no copyright, then it's in the public domain, so then I can go to these platforms. Not that I would, but if, if I wanted to, I can go to these platforms and go copy paste this prompt. Thank you. And the platforms can do nothing about it if it was written by AI. Uh, let me, let me twist the scenario. Go to this platform and look what it's doing, and then, uh, take the ideas from that and do it again. So, copying and pasting the expression that would be, uh, bad, but taking the ideas. You can't steal, you can't infringe ideas. Ideas. This is a, this is an area of fraught, fraught with risks and danger. Now let's go to a new scenario. Uh, Hey, Claude Code, look at the user interface in competitive interface. Yeah. Now take this interface and replicate it afresh. Yep, yep, yep. Take the ideas from this interface, vibe it out for me in React. Right. Take the ideas. Don't take any of the copyright expressions. Yep. Just take the ideas and then reconstitute it over here. Yep. Yep. Where's the moat of anything? Oh, what are we doing here, Damien? Let's just go, let's go live out the rest of our days in some, uh, farm somewhere. I, I jokingly, not jokingly share a link to a, uh, a property in Kiama. Kiama is a, a little town in Australia, about an hour and a half, two hours south of Sydney. Uh, and, and this property is kind of like on the left hand side of the highway with a backyard farm, overlooks the ocean. Look, it's a cool 4 million. Australian, Damien. Okay. Okay. Uh, you know, but money's not gonna matter. Money's not gonna matter anymore. Um, and, and, and I'm thinking, Hey, that, that's not a bad place to live out the rest of my days until the robots take over. And, well, and I've, I've heard that with the, if there's a nuclear war, then, uh, New Zealand and Australia wanting to take, we're too far to be hit. Yeah. Yeah, yeah, yeah. So, so as the, as the, you know, nuclear fallout happened, you, you can come and stay in my, uh, backyard farm. That'd be great. We, we could hang out with the kangaroos and wallabies. We, we, we jump into pouches and go shopping. Uh, I mean, there, there really is what, what. What are our children going to be doing? What are, what are we gonna be doing? Well, it's, it's a story from, from my friend in Australia, right? Like if the 90% drop is true, and, and depending on how long that goes for and which industries it affects. Yeah. How, how many pitchforks are gonna come out and how many, how much more populism is going to be there. I mean, there's, there's. I, I think we're in for some bumpy rides before, uh, we reach the utopia that you and I hope, yeah. In the one in 10 million. What was the number? Are you, are you a student in history? Uh, of course. Uh, I'm not as much. Okay. But I, I'm not a good student of history. Okay. Maybe I'm the same boat. Um, but I, I'm, I'm trying to think like the French Revolution, how many years did that go on before they kind of cross over to prosperity? That's a, I don't know the answer to that. Neither do I. And we should look that up afterwards. I, I'd like to think it's rapid, because, uh, like lots of heads got chopped. They were, yeah. And then, then that was that, right? Yeah. Yeah. Uh, so how many heads need to get chopped? Well, that's a grim way of, of, of ending the podcast in New York City. Well, let's, let's start with some optimism. I think that there's never been a better time to build. Oh, whoa, whoa, whoa. Okay. There's never been a better time to build, there's never been a harder time to be noticed. That's exactly right. You build, if you build something in the forest and no one's there to hear it. Yes. Does it really make a sound good at all? But, but, uh, I, I think that if we're gonna send some people, people off. Download Claude Code, build today, speak and will things into existence. And I think that you and your organization will be doing better. Yeah. I mean it's, it's no longer about, I think I wrote this last week. It's no longer, uh, an exercise of debugging. It's an exercise of managing, directing, and Claude will figure it out. That's right. You waste a lot of tokens. Oh, unless you use Opus 4.6 at high thinking and then, uh, then it uses more efficient tokens. And it's because it's more intelligent. We don't pay for that anyway. It's all part of our subscription. Right. Exactly. For now, for now, it's all subsidized. Right? Right. That's true. Like I, I think I told you like I was using API before I got the $200 a month subscription. I was doing API and, and I, I burned up $300 worth of API and uh, the first weekend, right. I thought maybe I should pay 200 bucks a month rather than $300 a weekend. That makes a lot more sense. It was a much, uh, much, uh, very good choice, but I, I think that we're, we're really going to be, um, you can build at the speed of thought. Yes. And you think fast and your team thinks fast. You think fast. Like how much more software is gonna be built really quickly and Yep. Single use, extremely fit for purpose. And I, I think it's a good time. I think it's a good time. My optimism having kind of walked around the floor in LegalWeek is, I think there's now a a lot more roundedness. In how software is made, in how software is pitched, in how people do buying decisions. Um, that wasn't the case last year. I think there were, there were a lot more starry-eyed admiration of the magic. I think this year is a lot more grounded on what really gets adopted. Um, I was having this conversation with a, an, an innovation person at a law firm, and he told me that he heard a story, again, hearsay, right? He heard a story. Where someone bought a purpose-built real estate piece of Gen AI kit. It's called Orbital. I I, so real estate specific gen AI platform orbital. They adopted it just for the real estate team. Once they rolled this out, what they found was the usage of that tool was way, way, way up, right? A very well considered purchase decision yielded an incredible outcome for the real estate team. A byproduct. A consequence of this that they didn't expect was the usage by the real estate team of Harvey dropped to nearly zero. Hmm. Displacement. Narrow versus broad. Yeah. And I think people are now looking, people are now being more considered, and I think that's gonna lead to better outcomes for everybody as long as the product solves something and the beneficiaries feel it. Uh, I know that we are, uh, trying to close out. Uh, let me, let me take, come go. But what, what you've just said reminded me of... um, so I think that we, the niche rifle shot that you just described, of course, of course. This is the way the software for one. Um, I'm also thinking about, um, because I vibe coded this thing in a weekend, I open sourced it. Yep. Right? And so because I opened, sourced it, you can download the thing that I did and you could say, I like it, but I think you should tweak it this way. And then you can make a fork of it and be able to do it that way. And then somebody in the organization said, that's fine for your department, but my department does it this way. And they make a fork be yours and you can fork. Turtle. No, I just do it this way for me and I can make a fork of it that way. So I wonder if yes, you have a rifle shot to compete with Harvey's to be able to do just your thing. And I wonder if the sunk cost or lack of sunk cost is so damn cheap to build the thing. I just open source it. Then that means that we could just infinitely fork and be able to make software, uh, for one. But I don't even have to build that software for one because somebody opened it open, sourced it. I just need to modify somewhere along the chain. You've just given every CIO a massive heart attack. But it's accurate, right? Open source eats the world. Yeah. Right. I mean, Linux runs so many things. Yep, yep. Because it's cheap and forcable and you can do whatever you want with it. So as code gets cheap, open source, it fork it, but then where does SaaS go? Yeah. And where does software go? Uh, I think there's still a place for well-crafted software that does something that is promised, that does it at a high enough quality. Uh, but yeah, it's a great question. The future is infinite. It, it is infinite. I, I was, I was trying to be optimistic saying you could fork things, but I worried that I was pessimistic. That maybe, maybe that might be sad. It, it, it depends on what happens. It depends on what happens and we don't know what's gonna happen. What we do know is there are some really good people in the industry who are building towards the best possible outcome. And if we all pull in that direction, I think we can get there. I'm glad that I'm sitting across the table with one of those people. Same. Cheers. Cheers, Ian. And thank you for listeners for sticking with us for two seasons. It's been great fun. I think, uh, this is, uh, the next season might be the best yet. Oh, every season's the best yet. Thanks everyone. Thanks everyone.