Transcription
Hey there, I'm Zach Bramitz and I am Legally Disrupted on episode 37 of Zack Bramitz's Legally Disrupted. I'm on another excellent legal AI adventure with my good pal Richard Trowman's from artificial lawyer.com. We're talking about some of the big news including uh Claude entering the legal market and uh as a result sinking the uh the market value of several publicly traded legal software stocks. So, we got some hot news. AI isn't slowing down. It's only moving faster. It's a good thing that I've got Richard along for the journey. I mean, adventure. So, let's get right into it. Let's get disrupted.
So, I wish I could tell you that I've like got my travel schedule like all down this year, know exactly which conferences I'm going to. Um, but I don't. So, I obviously know I will be at some of the Legal Innovator events for sure. That's a that's a constant for me. So tell me like what is what is the schedule this year >> kicking off in San Francisco again?
>> Yeah. So um hi everybody uh Richard again. Um so we'll be we'll be kicking off Legal Innovators in California as always for Legal Innovators California which is in San Francisco 2 days early June. Um check it out on the website. Then uh it's about a week off and then we are doing Legal Innovators Europe. It's in Paris but it's for Europe, right? Okay. So everybody in Europe, everybody who's developing legal tech, buying legal tech in Europe and people who are selling into Europe. So from the US, UK, Middle East, Australia, wherever. That's that's the basic idea. It's not a French legal tech conference. It's taking place in Paris because Paris is a beautiful city and lots of people like to go there. But yes, uh think of it as a European conference. So, we're going to talk about things like sovereign AI, which is a very big deal in Europe, and all the various things, but with a European flavor. Uh, then later in the year in November, we've got the London event and the New York event, the same as we did last year, which also will be uh I believe uh hopefully with La Watkins in New York and in London will be at the new venue as before.
>> Very cool. um as as we've been like, you know, get we're like, you know, already February, which is crazy to me. Um I I feel like I still haven't like uh really finished 2025 yet, but you know, January is always kind of like a warm-up month. Um what are what are some of the things that like you are deeply focused on, stories that you're tracking? um crit critical critical points that merit further discussion. A an excellent AI legal adventure, if you will.
>> Absolutely. Absolutely. Well, I mean, there's several things. Uh I mean, from the top, and it's literally I literally just posted it on artificial lawyer. This is this is Monday, February the 2nd, by the way, if you're wondering why I'm talking about something that happened a couple of weeks ago when this uh video gets published. But um this morning I artificial lawyer published the news that Anthropic is moving into legal tech via plugins for um a whole range of different areas for the enterprise. So it's doing like marketing, data analysis, but one of the verticals it's focusing on now is legal. Now, I don't know about you guys out there listening, but ever since I started Artificial Lawyer 10 years ago, we have always me and a whole bunch of other people have been wondering when will the big tech companies move into legal tech and they never did it. You know, every now and then, you know, Amazon would do something and everyone go, "Oh my god." Or Perplexity would do something and everyone be oh my god. And it but you know, it kind of got a bit of energy going, but then it just fizzled away. This, I think, is different. It's not massive. It's not a separate division. It's not like anthropic legal, right? It's nothing like that. We're very far from that. These are plugins and you're still going to have to use a tech team to help you utilize them. So, it's not that simple. However, as far as I understand, between OpenAI and Anthropic, this is the first time either of them have done something officially labeled legal, right? And if you look at the prompting, the instructions, the language, it's very sophisticated. This is not just, hey, review a contract. They go into it in great detail. Every single part, you know, like indemnities, liabilities, IP, data, blah, blah, blah, blah. There's a whole bunch of things. There's there's a set of half a dozen or more instructions on how the AI should think about redlining, just its approach in this context. And of course you can customize these. So I think that's really really interesting.
Um and then the question is who will it affect? Primarily it's going to affect the the most commoditized legal AI products. The people who are just who who people the people who've been getting away with it in terms of just selling very vanilla legal AI AI based products right you know airline based products. Um they still may do well. Well, I mean, you know, it amazes me how law firms will just buy anything with a word legal tech written in front of it, right? But I think increasingly the more sophisticated players will start to see the very high quality, super engineered, super strategic tools that do a lot that bring a lot to the table and all the ones that don't. And then I think they got a much better, you know, opportunity to do different things.
meaning you think you think that like you we'll see law firms by the end of the year who are working with like a like a base level of like an enterprise level uh
>> I mean look I mean Fresh Fresh Fields is already Fresh Fields has broken the mold already and he's doing this with um Google
>> right but what Google doesn't have is a dedicated legal division not a division of people but a division you might say of u tools for legal anthropic now does right and that's only going to grow, right? It will either grow and fizzle out because they don't get enough traction and the engineering and the investors, all those different parts of the the organization will just be like, "Yeah, it wasn't worth it. Kill it." Right? Or it does take off, does get some traction, starts to see some revenue bumps because of the increased usage by enterprise players and they go, "Wow, let's just let's let's accelerate this. Let's expand this." I think the other aspect and I you know I mentioned this at the end of the AL story about anthropic is that the the even new legal AI players have um have got two fronts now two flanks you know that are causing trouble. So you got your vibe coders which we were talking about off camera. You got you Jamie Sos and there's a whole bunch of other people there's a UK group as well doing it. um they're coming at one side and now you've got big AI and if if if open AI also starts to you know develop a sort of legal division as well that could add additional pressure well you know let's see what happens with the other players yeah and then and we should also mention you know Microsoft copilot because one of the things I'm finding and more and more law firms very often times ones that have subscriptions to a legal specific application um is they also have Microsoft copilot and I'm always surprised when I when I see that because to me it's like have one of them meaning like have a have a chat GPT enterprise you want to go with the you know the fresh fields approach and use Google you want to use like I I think one AI assistant for most law firms is probably enough what's surprising is how many of them have one or two plus
>> co-pilot I'm not saying that co-pilot's in the world on fire, but like you know it is there.
>> It's a job lot, isn't it? It comes with everything else. But I mean, I kind of see people who big up um co Microsoft copilot. It's almost code for we're not really using AI with any seriousness, you know, uh because what it means is that they're probably doing stuff like reddrafting emails and things like that. Doing very very light touch review, very surface level stuff. the people who are sort of serious about AI and the law are people who are bringing in the bigger, more complex, more high value tools. You know, for example, I'm not going to mention names because it sounds like
>> No, but I but but it's there are people out there who are coring all of their customers together. The customers are giving their data and they're using that to figure out what's market, right? You can't buy that on the open market. It doesn't exist on in open AI. Well, you can it'll go to Edgar, but you can't just use Edgar. It's not good enough, right? Or one of the very very very large platforms or a platform with tons of data. We all know who we're talking about, right? These are very special things, right? You can't just replace those.
Um, but you know, I think there's going to be there's going to be three group and and they may in fact, they probably will all happen at once. So, you're going to get your vibe coders who are going to be building customized tools. you're going to get the people who be going to be using stuff like this new anthropic legal addin uh for the commoditized stuff, right? The same way they might use the little calculator function on the bottom of their laptop.
>> Yeah.
>> And then there'll be the third group uh third area which will be the high quality tools. And I think all three things will be happening simultaneously.
See, I wouldn't call it the high quality tools. I I would call it like in other words like the legal uh legal applications right or spec or or you know specially designed legal applications because
>> well the thing is that you can you can see this thing is specific isn't it as well so it's like it's not it's not that they're specific so you can you can build a legal specific tool but it's very very very basic. I think it's it's the fact that there's a ton of engineering, there's a ton of data, you know, there's a there's a lot going on there that you can't easily replicate with a vibecoded little app. It's not the same. Here here's something I struggle with and I and I've I've seen uh Jamie discussing this and and I plan to discuss it with him as well, but vibe coding
>> part of the part of the attraction for my perspective and and why I'm I'm I'm bullish on vibe coding and you know this cuz I was at your conference, you know, both in London and in New York trying to explain to people that the most important deal of 2025 wasn't any investment Harvea, it was the Wix.com acquisition of base 44, right? That's that is the kind of like signal and the noise. That's like the the most important deal to focus on. And one of the reasons it's so important to focus on is because Bass 44 is a vibe coding application. And there are companies on that have been built on base 44 right now that are doing like $50,000 of of monthly ARR meaning it is a it is a true platform in the sense that it is creating other businesses that are that are gaining the value from it. So, you know, I've I've very much bullish on vibe coding
>> and where I think it's going to play a really really significant role moving forward is no, it's not like you can't like vibe code Harvey the the company or Harvey the platform. However, can you theoretically build applications that you yourself could use that may be better or more effective? No, they don't have to be scalable, but like who could they I mean like they're not going to be scalable like Harvey, but they don't they don't even necessarily have to be if all you're doing is using it for yourself. And I so I so I think one of the questions we're going to have is like you know like where the question in 2025 was do I need you know one of the specific legal AI applications for this use case or could I just use chat GPT right when like that like I think all the surveys that came out beginning of 2025 that were showing yeah like lawyers might be using AI but they're not necessarily using the tools that you identify as legal AI they're using the f the foundation models I think the question in 2026 is going to be can I use the solution that I personally vibe coded or do I need to use the firm approved tool for that? So I think that's going to be a brand new conversation.
>> Yeah, it'll be a bit of both. And the thing is is that why would you use a vibe coded tool? I mean, if you're doing it just to save money and you you work in a law firm that generates 4 billion a year in revenue, I'd say that was a false economy. You're you're putting billable time into something that is way more valuable well is is way less valuable than just doing your legal work. Um the question would be is it so unique this tool that you've built that none of the other platforms provide it? It's such a, you know, let's say you do super, you do you do these incredibly niche futures contracts. You do legal work for these. Yeah. Or, I don't know,
>> space rocket parts, something really, really, really esoteric, right?
>> And no one literally no one on the planet even has the data for it. You have the data. You are the you are the lawyer. You have all the contracts. It's it's a super bespoke thing. Yeah, go for it. V code it. Why not? Be crazy not to, right? You're not going to sit around and twiddle your thumbs waiting for a legal tech platform to build this thing for you, right? So, yeah, go for it. But how many cases are there for that? I think there are very few.
>> So, so let me but let me let me let me flip it around. So, I think there's you addressed like the value in terms of like is it less expensive? Now, if you are looking to implement in your legal department because someone so you saw Gartner told you that you have to. So, you're looking to implement a CLM, right? Do you buy a CLM where the likely price tag on that is somewhere between half a million to $2 million a year for getting the implementation,
>> right? And then you've got to take that off-the-shelf product and you've got to customize it to your specific workflow. And then anytime the workflow in the department changes, you've got to bring back the implementation partners and redo.
>> My question is is that the play or or at this point, do you just say, "Hey, listen, like we're we're going to we're going to constantly have a CLM and we're going to constantly iterate on it, but we're not going to like try to customize it. We will just literally rewrite the whole thing every time we change our workflow and it will we won't need it. I don't need to figure out like how do I make agileoft or isertis or ironclad work for me. I just build my own version because I know exactly what our department needs. So I think that's that that's a very very real case. But that's sort of just thinking about the money as the value.
>> Well, let's stay with the money. Let's stay with the money. Look the key thing is is that you know and this is this is as old as legal tech you know legal tech companies particularly well since 2015 2016 they came to market selling NLP products which obviously you have to train right and they basically wanted to sell software within about 6 weeks they realized oh no we have to provide consulting services right we have to literally get in there in person that's very expensive and then a whole range of consultants built up around that a whole ecosystem of consultants come in and help you with these AI tools and so forth. And many of them were traditional legal tech consultants who'd kind of upgraded themselves into this like new new field and that's going on with AI as well. So I guess for me the question is if there is an off the peg tool that really really quickly enables you to get to value even if it's relatively expensive. You could say that there's a lot of value there, right? there's, you know, you're cutting costs in other ways. You're cutting, you know, inefficiency and time and irritation. If the tool that needs to be bought off the peg is a gigantic pain to implement and you need tons of expensive consultants and it has to be updated every year and you you're basically constantly reaching out for help, like a car that constantly breaks down and you constantly need the engineers to come in and fix it, right? Like a race car, for example. If it's like that, then you could argue that the additional cost of building it yourself is negligible. But third point would be if your tool is so prone to breaking, so desperately in need of constant external input from a third party, it's not a great tool, is it? That's what I would say. It's rather like buying these, you know, like British sports cars, right? They look absolutely beautiful. They break down every six weeks, right? Or at least they used to. That was that was the old cliche, right? you know, Jaguar Jaguar E type. Beautiful car. Breaks down instantly. Same with the, you know, sort of Jensen's and other stuff, right? Buy a buy a Toyota. Okay, it's not amazing, but it will never break down. You just buy it and you just drive it until the wheels fall off, right? And I guess this is the point, isn't it? You if you buy a fancy solution and it's reliable, you could argue just stay with a fancy solution because you can just focus on using it then, which is of course what lawyers want to do. If it's a gigantic pain in the butt to operate and you constantly need third party help to keep it running, then you could argue, well, let's just skip to the chase and build it ourselves. So, I think that's that's the cost argument.
So the the implementation consultants is a a point I want to dwell on because I just I I have a um a little YouTube short about this from a previous uh podcast, but I I think that these implementation partners have been such a grift and I think in many cases like they are the ones who benefit from these implementations. They're long. They're drawn out. They're expensive. And so often these products don't even stand up and work inside the organization. And the one of the things, you know, I just recently, you know, you covered this, I invested in Sandstone, and one of the things that I like about them is there's no implementation partner at all.
>> It doesn't need an implementation partner. And I think that this the the implementation partners are kind of like a crutch that always keeps the product with a broken leg because the broken leg is really in the best incentives of the implementation partner. And uh and and I think that this is something that we should move away from. And as far as your other point that a lot of these, you know, previous implementation partners for various legal technology tools, whether it was a case management system or in matter management system or a CLM, essentially SQ SQL databases with some machine learning, you know, some basic machine learning built in. All of all of these products to me feel like they're deeply in trouble. And I think that that those those specialists that now move and say, "Oh, now I'm an AI implementation specialist." I I think what they they don't realize or maybe they do and they're just going to just kind of like brute force it. It's such a different um change management path. Meaning the the problem in AI, well the problem in CLM or CRM or any any of these types of tools was that if the employees don't use them, they're not particularly valuable. The problem is is that the employees using them bums out all the employees. They don't love using the products. They're clunky. They they I mean I I remember listening to a podcast explaining that like CLM only really works if you create a contract center of excellence within your organization of people who all they do all day is review contracts. And it's like wait if you have to like start changing around your department and the way you work and create these right and create these like contract like sweat facs like is the product really all that great? And so I like to me I think like the the with AI what the change management people haven't figured out is that like AI is awesome. It makes things better. It makes people's lives better. People like working this way better. And the and the challenge is not how do I convince them to use the AI. It's how do I convince them to use our AI as opposed to like whatever AI they happen to have privately and personally. And I think that Yeah. So I think so I I I I think that categorically these things are so different and I do not worry but I I I I raise a red flag for people who are going to bring in like the old you know standard consultants to to advise people on AI thinking that like oh yeah it's just like the last like decade of of technology products. It's not.
>> Well absolutely. I mean I mean fundamentally I mean legal tech um you know has been in gestation the gestation that we all wanted you might say you know the AI that we wanted but it's now here and you know you know the example that you gave earlier I mean if something needs tons of implementation and you have to force people to use it and you have to build your practice whether it's in-house or a law firm around the product I'd say you've failed and the product has failed because it's madness isn't it that's madness
>> you know I Obviously using technology does change your behavior, right? So we're talking, you know, through our laptops. Uh if you've got a car, that changes your transport behavior. So, you know, it's not it's not it's not irrelevant. Obviously, the technology does change your behavior, but it shouldn't completely alter it in a way that you start to do things that are inefficient. So, you know, you get that kind of like negative Jevrons paradox kind of scenario where the the efficiency actually starts to work against you. you you're so desperate to have this tool to to put it into action that you you have these other downstream effects that are negative that trickle through, right? Um, you know, like like you know, like driving to the end of the road when you could just walk, right? Exactly. And I I think this is going to be a big thing um especially this year in in in in legal technology, which is we're going to have to start separating between here's what the AI can do and here's a use case that's very valuable to me with AI. Meaning, I think an AI can draft an NDA. Is that a very valuable use case? Is it like very hard for you to draft an NDA? Typically, no, it's not. And and if I think there's going to be a variety of other things that we start to ask ourselves like, oh yes, it it can technically do this, but I think I might have a better way to do that one thing. And to your point, like I don't need a spaceship to cross the street. And I think we I think we need to start thinking about spaceship use case and less about crossing the street because I think we actually have very good ways to cross the street already.
Yeah. Well, no. Absolutely. Absolutely. The I mean and then and then you get to the whole thing about you know cognitive decline and all this kind of thing where people who are generally educated and intelligent get to the point where they can't write an email without jamming it through chat GPT. And you're just like what happened to your own judgment? You know how to write a good email. you know how to communicate with the client. You know, you don't have to put me through TGBT.
>> And and you know what, this is interesting because I think that there was like certain use cases that made sense in 2023 or 2024 that don't necessarily make sense today. And and and drafting an email is a perfect one. That was a great way to experiment early with AI and to and to like mimic the formality that we were spending so much time creating
>> and and start to wrap our brains around. Okay, well, wow, this is like something that I thought I would spend a lot of time on doing. But you know what's happened is the we our brains adjusted now. We know what an AI email looks like. So, you know what people are doing instead or what I recommend is be more human. It's okay to have an like a spelling error. You don't need to put bullet points in every single one of your emails. You can just kind of like grunt back at the other person and they're going to understand you better for in terms of like what it is you're trying to communicate as opposed to being so formal.
Yeah.
>> General AI output to me cuz I guess I write for a living and before that I mean I studied literature and etc etc. I mean for me it's absolutely it's like so transparent. It's like it's like glowing with fluorescent you know red glowing. I mean, if someone does a LinkedIn post or whatever, whatever it is, and they've used AI to to source some of the key information or to reshape it, it just screams at me and then and then I stop reading it because as soon as I think this is not authentic, this is not them. It's someone else's what was it's basically an amalgam of someone else's ideas which they've basically ploed and then rephrased for their own benefit. I just go, I'm not even really going to read this. And and and just for the record, artificial lawyer does not use AI in any shape or form in anything that I put out. Sometimes I do a bit of brainstorming very occasionally when I have a a thing that I don't understand or I need to get
>> right weekend. I might sit there and do some thought experiments. But Jesus, I would never write an article using AI. Never. like AI slop is, you know, okay, my eyes glaze over when I realize that you wrote this with AI, then I'm not I'm not interested. So that means that it mean doesn't mean you can you can sort of submit, you know, crappy content anymore, meaning the starting place is actually higher. But I think that like we now have to like rethink many of the use cases. And I think this is going to be very confusing because like there's so many partners and law firms who are like, I don't understand. You told me that like writing an email with AI is a great use case and now you're telling me it's like yeah, the stuff changes. It's one of the fastest moving like
>> people don't write don't write email if if you're trying to be authentic. Don't you I mean look I mean there's a huge difference between reading a 400page MSA or something and trying to find incredibly gnarly little red lines that you need to change to go off your playbook and then you want those to refer to past presidents and to what's market. Of course, you want to use AI for that because it speeds up your cognitive process, right? And in fact, is actually relying on your own mental output, but from previously, right? It's like it's like a charge on yourself.
>> There is a gig gigantic difference between doing something that it's kind of like a mix of art and mechanical,
>> right? There's a huge difference in that compared to someone literally just speaking off the cuff in an authentic voice in something that's not so structured like a news story for example. Yeah. Or a blog post that has to be real. It has to be authentic. Um, for me personally, because otherwise what's what's the point if if it's just regurgitated AI stuff, even if it reads well, you've you've effectively betrayed the spirit of what you're trying to do.
>> Yeah. You know what I mean? It's like a priest at a wedding using AI to do the to do the the ceremony. You just, oh, yeah, you could you maybe the AI is even better what this guy would have done, but it's so inauthentic. It's so unreal. It just feels wrong, right? But that's completely different. He's then going home and dealing with a parking ticket that he's got using AI because who cares if if it's his soul or not that's going into dealing with his parking ticket. Yes, absolutely. So I I I think we're we're going to move to a a a smarter kind of use case. And I think we're also going to and this is this sort of takes us to the last thing I wanted to talk about with you, which is Claudebot and
>> Maltbook and and that and that whole uh and that whole bag of beans.
>> But but the framing I want to start with is I think we're going to figure out that things that you can that you already do, right? If you're going to if your goal is, hey, I want to take the thing that I already do and I want AI to just replicate it and do it instead of me. I I think a lot of those use cases are going to fall by the wayside. I think instead what happens is we start thinking a lot more about like what could I not do beforehand and now with AI I could do
>> or or I think for it's like what what really leverages the value of AI. So what can AI do? AI can do all this creative stuff but I think that is a bad way to go the same way that you could buy a broomstick and run down the street bonking people on the end of it. Right? that you know doesn't mean just because you can do something doesn't mean that you should do that thing.
>> Correct.
>> Right.
>> The way to use AI is to leverage it to do things that enable you to be more you. Right? So it can go through all of your past documents. It can go to all your various playbooks and say look you know you've probably thought you're a very smart guy. You've probably thought about this already but if you hadn't have you considered using this terminology and you go oh okay thank you for that. And it will just say, "By the way, just in case you're wondering where this came from, this isn't AI slop. This actually came from your past three contracts that you wrote on a similar thing last year, right? Or you know, the person who sits next to you in the office wrote this and they don't mind you having it." You know, that is a sensible use because it because it zips straight to the value in a way that reinforces you being you, but you know, doesn't detract from you. Um, it's the for me it's it's and I think it's very very finely balanced and I think a lot of people have got lost you know like on the bottom of every LinkedIn you know post or when you're trying to post it says do you want to rewrite this with AI it's like no I do not want to rewrite it with AI what you know what what the hell are you trying to push on me
>> you know it's like it's like it's like hey would you like to drink some bleach
>> so so this is so this is why I like the Claudebot and molt book experiment even if Claudebot and molt book, you know, uh, end up not necessarily being the most important trend to follow moving forward. The reason I think it's such a important like mental frame to understand and internalize is because it starts getting into the world of like this is this is beyond what we could have imagined. This is this is this is truly truly groundbreaking. This isn't just, you know, the ability to write a letter in the voice of Shakespeare, right? or uh this this is something where it's like wait a minute there's it what first starts off as I'm going to download you know this uh this software this open source software on my computer and just give it access to everything
>> and then what the what the then like respon what what the sort of the next logical step of that is and then molt book which is an a a social network designed for AI agents and but that the truth is the truth is that that is to use a bit of Anglo-Saxon complete bollocks because not what you've said but what what they're trying what they're saying publicly because the truth is is that it's not a social network because these things do not have true identity they're relatively small agents that have been given persona right you know yeah you behave like this you will try and do this you will try and do that you will
>> oh well there are there are there are there are like sub molts I guess what that's what they call them where where they are where the AI is discussing is this a simulated experience or is this real
>> meaning again but see this is I mean effectively what is the guy the guy who created very clever but it's like a it's like a sort of um uh play your own adventure what have those called you know where you you buy it's like Dungeons and Dragons and you open the book and you you turn to the page
>> role playing
>> yeah it's like a role play playing game the whole thing but but it is just a role play. It's like, hey, here's a conceit that we're all going to share. Uh, let's pretend that this is going to build Skynet and all of these agents are really, really clever, blah, blah, blah, blah, blah. But they're not. They're just they're just simulated personas. They're relatively limited. In many cases, they're designed to say and do sensational things, which then people will screenshot, which then attracts more and becomes this bubble. And like you know as I explored in the you know artificial law article this morning about about old book I think for me it's a kind of a flawed experiment but the experiment hopefully will get people to think about using multiple and different uh agents together. So like one little example I gave was
>> what about solving a problem effectively by replicating all of the processes and personas inside a law firm. So you got your paralegals, your junior associates, your supervising probably a junior associate or a salaried partner, then the partner and then a bot, you know, an agent representing the client's needs. And you put all these together, but it only works in a meaningful way if you have all the data, you have a workflows, you have checks and balances, you have quality checking, otherwise you're just going to get like a whole load of rubbish.
>> But but you're saying it does it does sort of create that like, you know, house on the hill that that uh that's something that it's like, in other words, this is what's possible.
>> Exactly. It makes it makes you think and also as again as I pointed out in the article multiple agents working together there are legal tech companies who've already done this. So for example norm AI launched last year on the idea that they would create their own agents which would go into compliance kind of I don't know spaces inside banks and wherever and deal with the agents working there and flank for example has has already also looked at working with multiple agents and so have other companies. So the idea of multiple agents interacting with a net positive outcome already exists and I think we need to look at mold book as an inspiration to do something more sensible rather than seeing mold book as an end in itself.
>> Yeah.
>> So I I and I just want to add one one one thought on there uh which is I I've written a lot before about AI 1.0
>> and eventually moving to to 2.0 know and and the the example I gave from the internet is, you know,
>> CNN.com was what I consider
>> internet 1.0. Now, obviously hugely influential and cemented probably cemented CNN and ESPN, many other publishers as as relevant for it probably brought them, you know, 30 to 50 years of added relevance from investing in in early internet. But effectively what it was was just like the news that you get on the TV now you get it on the internet, right? It's it's not really all that different. Now compare that to Airbnb or Uber where I push a button on a phone and then something happens in the real world. A car shows up or I now can rent someone's house. I I like these are things we just would never have contemplated in the early days of the internet and it took a number of other developments materializing until we got there. But I think more than anything it was just an issue of like creativity and trust and I think that those two creativity and trust is how we get to AI 2.0. when we start trusting the AI a little bit more, okay, in the in in the broadest sense and creativity is being able to say like, okay, well, now if that's the way the world is going to be, then this is our response. And and I think that this is where the where you're going to get winners not just in startups but you're going to get winners in a lot of the service providers and incumbents because the law firm that recognizes okay this is how the world is moving and this is going to create a kind of really significant behavior change and this is the way that we need to respond and it it's not just like buying a tool. It's like what is how does how does our firm, you know, become maybe even a little bit more human in in in this in this uh in in this new in this new world. So, I think that molt book is a great thing for lawyers to learn about just to go like whoa, okay, that's like way different than like we were talking before like AI can draft an NDA. It's like yeah, AI can draft an NDA and AI can also create its own social network where agents can go and hang out and talk to each other, right? Like for lawyers, I think stop focusing and anyone for that matter, stop focusing so much on the like can it do my work better than me version and ask like how will this thing change the world and then how do I need to respond and adapt?
>> That to me is where the winners are going to be.
>> And I think that connects directly to a question that I've spent a lot of time thinking about which is where do we how do we triage this? How do we automate? How do we decide what to automate and what not? Because ultimately that's what we're talking about. Whether it's an agent or whatever it is, we're talking about automation. I you're handing over the performance of a task to something else, right? But it's not you
>> um or another person. The it's a tricky one. So for example, you know, I mean there are already automated things that you know, corporates do, you know. So if someone buys something, it it triggers a thing in an inventory system which then goes to another part of the uh you know the digital infrastructure of a company orders you know 2,000 more of them because it can feel that it there's going to be a hole soon
>> right in the you know the Mars bars whatever sbury supermarket so it goes right boom we need some more of those how do we how do we take this forward with with a gentic um AI so you know if you're a law firm or an in-house legal team how do you say, "Well, you know what? We just don't do NTAs anymore unless they have been flagged as a high risk or high complexity. How do we literally hand over work?" But then the the huge enormous $1 trillion caveat to all of that is accuracy. And I think it's interesting actually, I'm I'm definitely going to write about this again, is that the discussion around accuracy has gone away, which is kind of bizarre because it's in in a funny way the more
>> it hasn't been solved.
>> It hasn't been solved at all. I mean can you know I mean depends who you talk to and of of course accuracy itself is a very subjective term because there's a thousand different ways of considering what what does accuracy mean many many different ways of applying AI to different tasks and so forth but fundamentally you can't really hand over anything to agents unless you trust it going to your point where does trust come from trust come from accuracy right you know if you hire a cleaner for your home
>> y
>> and They always do a bad job. Even if they done 80% of the kitchen, right? You're going to fire them because you can't trust them because you keep saying, "Can you please clean the sink?" And they Yeah. Yeah. Yeah. They never do. Right. And you're just like, "What is the point in having this cleaner? They never do a good job. I have to do it myself." Right? Exactly. Same with agents. You've got to be able to trust them. You can only trust them if they're accurate. We're not at accuracy yet. So, that's really the the Achilles heel of all of this stuff. I mean, it's very very cool to have a bunch of agents talking about let's create a new religion right and then they're in turn tapping anthropics a huge sort of store of literature and then effectively regurgitating it in a sort of you know stockic um parrot type way but if you say something like you know write me an NDA and it screws it up because it's it's understanding of what you want is inaccurate and you've automated that, you've handed that over. You've effectively created a a weapon of of mass destruction within your own business because it's just pumping out wrong stuff and that is a fundamental problem. So, I think agents hopefully this will reignite um the debates around accuracy.
>> Richard, till our next great excellent legal AI adventure.
>> Thanks, Zach. It's a pleasure.