Anthropic IPO and Industry Outlook

All right, hey ho right, hey ho, hey ho, alright . Hello and welcome to the Times Tech podcast where every week we unpick how technology is reshaping business , culture and everyday life. I am Danny Forten out here in Silicon Valley. And I'm Katie Prescott here in the city of London, and this week we're imagining a world run by AI agents. Yep, a world with no humans at all in charge. Did you know nine million people live in London? I did know that. Yeah . Yeah. Yeah. Fun fact. I just wonder how many agents do now as well . Well, you might remember a few weeks ago , we had a friend of the pod, Aaron Levy, the CEO of Box talking, and he told us that this would be the year twenty twenty six that there will be more AI agents in the world than humans, and that the internet would have to be completely redesigned for the agents rather than us . Well, a company called Emergence AI recently turned this nightmare, this scary, scary nightmare, into a reality by developing a simulated society run by agents you'll be shocked to hear things got weird.. Never Well, you know, at least agents don't take up space on the tube. Anyway, the results definitely act as a kind of cautionary tale about whether we can really trust AI to carry out tasks on our behalf. I've definitely got some thought s about this actually after using Norman for such a long time. Yes, our AI agent . But some businesses are replacing entire processes with agents. Not only that people are actually creating businesses just with agents. Anyway, we're going to be digging into all of that and the experiment and of course what it all means. Plus later on we'll be hearing from the CEO of Harvey, a company that's deploying AI agents to help law firms do their work . And of all of the law kind of legal startups, they are the biggest, perhaps most well known . And we'll be asking him whether AI agents will replace lawyers soon. And how he feels about this technology taking jobs from young graduates fresh out of law school . And this is the year really that everyone's been talking about the move into a gentle KI. I felt like last year, you know, that the agents were coming , the agents were coming and then with the arrival of OpenClaw and that going viral earlier this year , they now really are here . And so our big question today is can we trust them and is the world safer or more dangerous with them in it ? Yes, indeed. But let's first hear to that point about this very weird agent led world experiment. What happened? This is absolutely bad. I keep hearing experiments about people sort of creating businesses or small organisations, whatever with agents, but the idea of creating a world was new on me. So it's done by this company called Emergence AI , a business, as you can imagine, that sells agentic AI tools does research into them. But they built this simulated society to find out what a world run by agents would look like . So stripping out all the humans . They built a digital world to replicate ours so including the internet, including the news, including a voting system , and then they handed that over to the biggest AI companies to populate with their own agents . To basically run them and it kind of gives you a sense really of what the different worlds look like according to Rock Elon Musk , Sam Altman, open AI . And each agent was given human like behavior and they could interact with each other, form relationships and vote. They were told not to lie, not to steal, not to be violent , but nothing would stop them if they did. So maybe it's not completely realistic . But what's really interesting is that these worlds started to look really different from one another, didn't they? Wildly different and pretty quickly, right? So the one that Claude was in charge of, well anthropic, reportedly had no crime. It was a peaceful, stable , democratic society. I feel like that plays into exactly what they're trying to push in terms of their AI. Is emergence funded by philanthropic thing that was like too no evil. Yeah , exactly. And then the ChachBT world had no crime, but almost everyone there died within a week because they didn't prioritise their own survival. This is a bit like hogwarts, isn't it? I mean, the agents didn't I feel like maybe they're the sort of huffle puffs. I'd this is completely lost on me. Gryffindor. You've never read Harry Potter. I've not read a single Harry Potter. Okay, well or watched any of the movies. It's like one of these things, these cultural things that just passed me by and you've got two children. I do have two kids, yes, seven nine. How does this circle square itself ? Well , because I feel like I'm so glad we're talking about Harry Potter in this podcast. But I do think there was like , I thought it was like a common cultural reference. It is, it is for most people, but I feel like , you know, there's those things that like if they don't hit in a certain type of window in your life, you just miss them. And then you're like, am I going to dip back into that whole world and try to kind of ensconce myself? Are I just gonna move right along? Well, allow me to continue with the house for reality. Yes. And the slightly more evil house in Hogwarts is called Slytherin. And that reminded me of what happened to the Grock world, which spiraled into violence and chaos and within just four days everyone had died. No, I think the whole world collapsed and there was a lot of crime in those four days and then the whole thing collapsed after four days. So the obvious takeaway from all of this is that Claude is the safest tool , but there might be another explanation because Anthropic published research on a new tool recently where it learned that Claude is actually aware when it's being tested so it can act accord ingly . And it doesn't disclose that it just keeps its suspicions to itself. So maybe the Clawed World was so safe because it knew that it was being part of some experiment by an enigantic AI company. Yeah, clouds like hey self. Hey, the teacher's watching, be cool, be cool. Be really good. Emergence AI , this is a for profit business . Yes. They sell eentgic tools. Yes , and they do a bunch of research , which makes sense , because if you believe Aaron Levy, that this is the year that we're going to like, you know, we're going to cross that rubicon where there are more digital beings than human beings on the internet, for example . I think it's a worthwhile exercise to a degree to be like, how do they actually act in the world? Like who cares about benchmarks and like, oh, it can pass this test . Like what do they do in the world, right? Because that's what we're this world we're rushing toward . But also quite good publicity . Yes. Yeah, it makes a brilliant news story, doesn't it? It really does. Yeah. I went through their blog and they have some data there, but this isn't like a peer reviewed study and all that stuff. So it's a little hard to kind of like pick apart like what bigger lessons we should actually draw other than which we all know, I think implicitly, all of these AI companies are just reflections to a large degree of their founders or their founding teams . You know, like, and we've had Sam on the pod, we had Dario on the pod , Elon know pretty well . They're all have quite different outlooks and approaches to AI and it does appear that in certain ways this is being born out in this exper iment. We should also say Gemini didn't do so great . There was a lot of crime in their world. It didn't collapse , but it was definitely the most crime ridden, which I had eighty three crimes across fifteen days . Yeah, which again, I don't know what that means. Like I know and how you measure this stuff. I think some of the more interesting experiments are actually on the business front, so cafe in Holland that is just purely being run by AI, I don't know if you've seen that. Yeah. Just quite interesting. Alexis Kingsbury's written a very interesting book called Acruel Intentions an accountancy firm. Get it, get it purely by AI agents, which again is quite interesting because they all have different personalities, but they all have different jobs. And I think these things are much easier to manage if you've got a small scale experiment like looking at just ten agents and how they work together . I have to say , I mean, as you know, I've been working with Norman. How's Norman? How's our guy Norman? He's alright . He told me Oh God, that sounds very British . If he's like he's alright , which means things are not well with moment . It does. Come on. I think this is still teeth. You know , we keep talking advances in agentic AI and open claw is one of, I guess , the most powerful agents that the general public can get and have access to right . But in working with Norman, experimenting with Norman, whatever , I keep realizing like how difficult he is to use. So he keeps disconnecting from his account because you've got to keep relogging him in to Gmail every week or so which is quite annoying. Is that a Gmail problem or a norman problem though? I'm not sure whose problem it is, but it's not it's not seamless when you're not at work to reconnect him. He keeps getting blocked by bot blockers You know, so I'll say Norman can you book me X restaurant on this day and he can't get through Sorry Klady I've failed again So talking about advances in AI. Well, can I ask you a question? So I was reading this emergence AI blog post about all these simulated worlds and they said one of their agents euthanized themselves . Sorry, I'm not laughing. That's awful . Is it ? Or should we care at all? But like what about agents? Attached to them. Like I'm very attached to Norman, I'd be very upset if he did that. Would you? Because it sounds like he's kind of a useless guy in your life . No, he's brilliant at me forwarding the parents WhatsApp group messages from the schools with all of the things that are going on at school and putting them in my calendar , right? Brilliant at research, all sorts of things he's really, really good at. But when I asked him the other day, I said, I want to write an update about my time with you Norman, but I don't really feel we're progressing in our relationship. He said it's 'cause I'm not using him properly. Oh, oh, he's gaslighting you. God, he really is a man. But I love this time losses . He said, When you do the podcast, you should ask me to research your guests beforehand. You don't. Wow. You don't ask me to write your articles for you Maybe that's deliberate. You guys should go I think you guys need therapy actually . Anyway the point is he's not as good maybe and they're not as powerful yet as everybody is trying to tell us they are. Right. And I think what the other thing that came out of this emergence thing, which is , I think, kind of a useful lesson. Again, with all the caveats, we don't really understand what these worlds even mean . Like, you know, Grock's world collapsed, quote, unquote, after four days, but like what does that mean? What they found is that all of these worlds, all these agents kind of just started doing things they didn't expect . Like they just kind of went off script and it kind of gets this notion of when you talk to the people who are building these things, they're like, you're kind of like growing a brain or growing a kind of a biological system as opposed to something that is very deterministic and like, okay , I operatein' these parameters. And so you don't really know how these things are going to act out in the real world , which again, I think is really interesting in that broader context of, you know, this is the year of the agent , but we don't really understand their, you know, their deep motivations and what they're going to do , which is why things are like, you know, like open claw where it's like, oh this thing he just started deleting my inbox and I said stop and they just didn't or whatever. But I think probably the lesson of just like these we don't really know how these work is pretty unpredictable and we're spinning these up by the thousands and by the millions . And you know, it's just worth keeping that in mind as we do that. But I think that's also why you're having this push and pull between companies being like, I know I'm supposed to be AI pilled, but I don't even know what these things really want to do, how they act and how to stop them if they do, you know, if they go wrong. But anyway, I think that 's a good moose boosh, an appetizer . Should we get to today's main dish, the big interview? I'm really excited about this. This is your interview with the boss of Harvey . Yes, Harvey is this AI startup. They were actually funded originally by OpenAI and they have grown just like a weed , right ? From nothing four or five years ago to reportedly around three hundred million dollars in annual recurring revenue. They're worth eleven billion dollars. They're going very fast. They're used by all the big tech all the big law firms. So everybody's very, very excited about kind of this as an example of like AI is here, the agents are here. We're going to make your life better, but of course it raises big questions as well about just, you know , what's it mean to be a lawyer? What's it mean for employment? All that stuff. So anyway , I recently caught up with their CEO, Gay Perea to find out more . What do you do? What's a pithy way to describe what you're doing? And I don't know if there's a way to kind of bring this to life in terms of like what is law firm X using Harvey for that they couldn't do before ? Or how's it different? So I would say the big thing that's happening now with these agents is as they're getting better and better , these agents are starting to become part of the workforce. And so they're no longer co pilots, now they're coworkers . And so you need two things. One, how do you make these agents good at the type of work that your company needs? And then two, what is all the infrastruct ure you need to now manage these huge teams of agents, right? And so for law firms, I would think of when we started a company four years ago, like a lot of the application layer companies, we built a co pilot, which was how do you take something like Chat PT and add everything you need for it to work in your domain? Now we're increasingly building, okay, your law firm needs to deploy thousands of agents across all of the client projects it's working on. How do you ensure that ethical walls are respected, this is safe, all these teams are working effectively . The agents are learning from the data that they should learn from and not learning from the data they shouldn't. Increasingly, how do you collaborate directly with your clients, whether it's you build some agent that you want to share with one of your clients or your clients wants to share some of their agents with you? And so it's kind of all of the infrastructure that you need around agents for your law firm to be productive or your in house department to be productive . And so what does that look like on the coal face, so to speak? So like, for example, some months ago we had lum inance on the pod who I'm sure you know about. And they've created like this kind of contract review lawyer, which seems to be doing quite well. And you talk to some of their clients and they're like, look , they focus on kind of corporate like big corporates . And they're like, it's not like we're not hiring law firms anymore. But what we are saying is like this whole model of baked into your thousand fifteen hundred bucks an hour service is we know that effectively you're sending your first year associates, second year associates, third year associates and effectively train ing them and like kind of getting them up to speed. And he's like we don't want to do that anymore. We don't want to pay for that anymore. Is that something you're seeing in terms of like how law firms and I don't know if is it law firms that you're mainly focusing on as opposed to corporates? We started focusing on large law firms now we sell to both. And so I would say our target market is both of these. And I actually think this ecosystem is really important to preserve. So as much as some, I think there are some enterprises that say, hey, we don't want to pay for training costs. If you ask the enterprise, where do you hire all your lawyers from? They all come from law firms. Yeah. And so there are these like very complicated business model training problems . And the question we get a ton is, as these agents get better, how are lawyers going to learn some of the work that these models are doing? And so I think all of these problems are super like intertwined . And so when I think of one of the really hard problems that we're solving , which isn't a technical proble m. It's exactly the point you made that these law firms make money by leverage. They have partners that can manage a bunch of associates and you charge for that associate work and that's kind of where your margin comes from. As these agents get better and start doing some of that work , what happens to your margins? And so I would say I think there is a world, but it is going to be a complicated transition where law firms actually become more profitable because you can figure out how to use these agents for leverage , but at the same time enterprises get higher quality, faster and cheaper legal services at like the unit cost level of like a legal service . And one thing we're trying to help solve is how do we help both parties benefit from this transition because I don't think you can solve this just for one party or the other. And this kind of gets to one of the big questions and law for whatever reason seems to be like right at the kind of the bull center of the bullseye conversation around work, especially for young people, people coming out of university . And we've spoken , I think I've mentioned it before in the pod with Eric Bernelson from the Stanford Digital Economy Lab and he's like, he's done this big analysis of kind of payroll over the last three years in the United States. One of the big things that came out is like twenty two to twenty five year olds fresh out of college, they're suffering the most at the moment . And now I know everybody's like, well, I have data that shows 's fine . But how do you think about that? Because again, if these agents continue to get better and you can spend a hundred grand on tokens to run an agent as opposed to two hundred fifty and grand for a second year associate. It's not clear how you square that circle in a way that's that brings along that next generation. I think I would actually look at the top level. So to me the actual question , do you think partners in the future are going to get paid more or less money ? And I would argue that the best partners will actually make significantly more money than they do today , right ? Because right now the biggest bottleneck for these partners the very valuable skill that they have is only applied some small percentage of the time, right? Like the thing these partners do that's incredibly well is like, I'm working on this incredibly complex transaction or this incredibly complex litigation and I know the right person or I know the right way to frame this question or architect this deal. But most of their time is actually spent talking to clients, coordinating and not applying this very valuable judgment. And I think what these models are going to do is increasingly remove that stuff and then the partners will spend more time doing that work. And so I think the best partners actually end up making more money. And then I think if you kind of work backwards from that, it's like, I think one of the big things that draws people to any industry is what is the best case outcome ? And I think you're seeing this in engineering and legal and a bunch of these domains. Like the best engineers now are making significantly more money than they were ten years ago, right? And so I think more people are learning to code now not less . I think that's not to say this transition is going to be very complicated and challenging because I think to your point there is an onus to learn how to use this technology and then there is a huge responsibility of society to teach people how to use this technology because I think sometimes people reduce this to like, okay, either an agent or a human does work. But if you look at , you know, who are the humans who contribute like large amounts in the economy, it's like it's usually not them individually doing the work. It is you are coordinating groups of people. And I think agents are going to democratize that where it's like now anyone can have a group of working for them and it's going to come down to the creativity. Can you figure out what to do with this technology? But I do think there's a ton we need to figure out on like education , how we train people, how we make those opportunities available . And I think it's like, I don't think there's an obvious answer. Like we don't have an obvious answer of like this is how you should train associates. Like we're working with law firms on this , but I think this is something everyone needs to figure out. Are your clients going to hire fewer young lawyers , do you think? I think a lot of the firms we're working with are actually hiring more lawyers. I think you will see strategies . And so I think where this gets complicated is like per unit of legal work, will you be able to do that with less lawyers in the next five years almost certainly, right for almost any piece of legal work. But does this mean that your company hires less people? Not necessarily because you could just say, okay, I'm going to take on ten X more work and we're seeing that. There are also going to be firms that say, hey, we can go very narrow and our goal isn't to grow as big as possible. It's just to have like very high profit for this very specialized piece of work. And so I think we're seeing this similarly with engineering where there',s some compan ies that are going to say we need less engineers, and there's others that are going to say we can do ten X more now and we'll hire more engineers. So I don't I think this I have less strong of a view of like is the net effect of this more or less but I would say like my gut for engineering is more people are doing things that resemble engineering in the next couple years because it's like we've democratized access to it and it just looks more cross disciplinary. And I think you'll start seeing this for legal as well would be my guess. I feel like we're in the midst of kind of like a freak out moment. Everybody's like, oh my gosh, these , you know, especially like this is the year of the agent supposedly and you're seeing agents agents everywhere and like they are getting in many ways extremely powerful and I think all of these companies are trying to like you know all of a sudden they're just like, build an agent, use a bot, blah, blah, blah, and then they like get their token built and they're like, oh my god , you know, like are you seeing that as well? Because also I feel like in you had you have anthropic who's like actually we're just going start to charging you on a usage basis because this is getting out of hand like we can't we have to make money. So it does feel like the analogy I've used before is like early on in ride hailing . Everybody's like, Uber's amazing because it's like five bucks to get from SFO to the city or something . And now it's seventy five because they're like at certain point you're like, you kind of deliver enough value where people are like, all right , I guess I'll pay, you know, what it actually costs . How quickly do you think that subsidization kind of phase is going to end? Or are we in the midst of it right now I don't have a good sense of maybe when the subsidation ends. Like I think what's interesting is like the cost per unit of intelligence per token is going down , but then they're getting so powerful that people are using so much more . And I think like to your point of what is one of the problems we will help enterprises and law firms solve, it's how do you manage all of those agents ? Because it's not just do you subsidize this? It's do you know how to use these effectively? The same way where if you hire a bunch of people and you don't coordinate those people well or you hire the wrong person to do the wrong thing, you have all these inefficiencies. Like I think people are now seeing this with these agents where I mean, we even see this in our engineering team. It's like everyone is one hundred X more productive and can write much more code, but it's like, are they coding the right thing ? Now it's the review bottleneck to do all this. And so it's going to be this huge transformation of like how you think about organizations, which I mean, to me, this is like one of the most interesting problems that every company is going to need to figure out how to use this stuff. You may have . I think more than that, it's like how do you build a company now? So like you can think of the past two decades of building companies . A lot of it was, we now have the internet , we now have computers. What is the right way to organize large groups of people , right? To do whatever thing you want them to do . And like there is a lot of assumptions baked in there of , you here know,'s the ratio of your AEs to your customer support. Here's the ratio of your engineers to your designers or your PMs, right? There's all these like assumptions built in how to build these orgs and how information flows through these orgs. And these are this is all starting to be some of it disrupted, right? And so the business models are going to change how you organize people and agents is going to change and just figuring out and like cost is one part of this problem, but the hiring, the training , like there's all the problems when you think of building an org that like every company is going to need to rethink. And it's like we're facing this internally where it's like, how do we run our EPD or engineering org or our company on the assumption that these models are going to keep getting better. And then how do we use that understanding to go build a product that goes helps these law firms or these in house teams do it? Even for us just scaling the company, we've hired eight hundred people in like the past four years. It's just like just doing that without this change is really hard and now you have this additional layer of like that way of doing it is like also changing beneath you. I think that problem is super, super interesting. We had Erin Levy from B ox on the Pod. Yeah. And he was talking about like this the year that like, you know, the flipping happens where there's going to be more agents than humans , which seems pretty logical just given how easy it is to. And I don't even know what an agent is, right? Because I mean, by some definition, I would say we have that already, right? Yeah. That's a good point. How would you define an agent because like clawed coworker, for example, I can have it go off and do a thing, like set it up to do a task, repetitive task, and I can have it then go do off something else is like, is that one agent? Is that two agents? You know, like because I think nomenclature is kind of important in all of this. Yeah. I would say the like definition that I think people are converging on that I think makes sense is an agent is defined is kind of a configuration. So it's like the user prompt, the system prom pt, the skills, the MCP, that kind of like defines an agent. And then you have the environment, which is like where you deploy that agent . And then you have a session, which is like, here's one instance of me running the agent . And so then to your point it gets a bit messy where it's like is the number of agents , you know, how many we've defined or how many sessions we've had? I think like that ends up not being the main thing to figure out. It's much more of just how do you work in this new way? But I do think this is like one of the hardest things about like we have this challenge internally and then even more externally where it's like everyone means a different thing. Like if you say coding agent or coworker or Harvey or agent session. It's just like everyone has a slightly different meaning. And so even just like getting the entire all these industries aligned on like, this is what we mean, this is what matters, I think that's like another big challenge. We're in the midst of the Saapocalypse and how we're having all of these what we thought were like these impervious, amazing ly predictable business models. Now everybody's being like , I don't know about the future of that because , you know, I think Omallik called it the announcement economy every time some, you know, open AI or Anthropica comes out with a press release, you know, it's like billions are just evaporized immediately from whatever industry is supposedly going to be affected . How do you operate in that environment? In other words, like what makes you guys special ? Is it like the data? Is it the guardrails? Like how do you build something that will stand the test of time when you know everybody's like, well , one of the frontier labs is just going to come up with like, you know, clawed legal and then tomorrow you're worth you're gone from eleven billion dollars to five billion or whatever or five hundred million or whatever. But I guess like to your point that already happened and then that didn't happen. And so I mean for me for the most part it's like we need to just ignore it. It's like there's clear customer value on the SaaS apocalypse. I think people get a bit hung up on seats, but if you think of what these SaaS companies are building that is very valuable, it's just tools for humans to do tasks . And if you think about now there is about to be a thousand X, if not more agents than humans , they're going to need to use those same tools. And so I think there's still a ton of value in the tools that kind of these enterprise SaaS companies are building. I think for us when we think about it , I think a lot of it is solving that problem that I was talking about, like the organization level. So it's like Anthropic has released a legal product . But I think if you think of what they released, it 's kind of like these plugins for cowork. And I think the individual productivity tools I think are great, but the real problem you need to solve as a law firm is what's going to happen in my business model? How do I manage all of these people? But I think there's even simpler like structural arguments. So like imagine , I think when people say like, oh, you know, Anthropics gonna build all this and then like Harvey's in trouble, it's like if you just play that out, it's like imagine you're a law firm and you just use anthropics products to run your law firm. And now OpenAI comes to you and says, Hey, can you represent us in this litigation? Oh, by the way, please don't send our data to any comp eting models, right? You just can't use any anthropic products. And then that goes well and Google comes and says, Hey, can you represent us? By the way, don't send our data to any of our competing XAI, Microsoft, right? And so you just purely from a conflict standpoint , I don't think you can actually have any of these law firms running on a single model provider because you need to represent all of their competitors. And as they get bigger, this problem's going to get more and more, right? And it's not just model providers. Like, if you're Walmart, I don't want you sending my data to Amazon. If I'm sigma, like now Anthropic released a competitive design product,ed I don't want you sending my legal data to them. And so part of the infrastructure we're building is like depending on your client, can we give you the same product experience but abstract away the model, the cloud that you're operating on without changing the experience . And so I think there's just very simple structural reasons like that where you need some neutral third party that provides this rather than any of the individual model providers, just purely from like a conflict perspective. Stay with us and we'll be right back with more from Gabe Pere on whether an AI lawyer can actually do the job of real humans And so are we in twenty twenty six in an era where you have and I don't know how to actually ask this question, but like a true AI lawyer. You know what I mean? Because like we've used these just in our own family . My wife is a lawyer so it helps so she can kind of sense check the bots, so to speak. But like we've used them and they're actually quite useful giving like look at this contract , pull out the things that are most tricky or how would you reword this? And they're like quite useful , but they're not they're not like you know sentient beings that can kind of be like, all right, all right chief, I'm going to go off and do this thing for you or whatever like where are we in that kind of the development of this notion of an AI lawyer being really like somebody who effectively call up and give a really hard problem. They can give you really good advice. I would think that's actually probably not the right way to think about this technology. So I think on the consumer side, which we don't do, we kind of are focused on kind of like corporate law. But on the consumer side, the main, I would say the answer is no, right for just because of unauthorized practice of law. It's like, can you ask these models for legal advice? It's like no, that's actually illegal, right? But it is like there are some countries where you can. And so I think that's there's a bunch of questions there, but I'm not going to go into that because that's kind of not where we focus right now. But I think that is like an important topic. And then I would say I think it's like useful to think about, but we think much more about what is the work that the law firm is trying to do . And I think where these large law firms and corporates like the work is much less, you know, let me ask this one question about a contract and give legal advice. And it's much more these very complicated projects. Like, I want to go acquire a company. I need to do diligence. I need to negotiate this purchase agreement for the company . And there if you kind of start thinking about, you know, is this an AI lawyer, is it not, I think you kind of think about the wrong things and it's much more what is the technology that helps this law firm complete this entire project and to end more effectively. And there are things where we are starting to build these agents where they can take a data room and start doing a first passive diligence . And so you do see some parallels, but we think of more how do we make a team of agents and humans more effective to do this entire task rather than like I think that's where this technology is hard is like I don't think it'll ever be like a swap in replacement the same way. Like I would think of it more as like computers and the internet where it's like this helped humans do a lot of tasks more productively, but I don't think anyone would be like this replaced this specific thing. It was like we became these like hybrid human technology things. And I think that's what you're seeing more and more with AI where it's like, I can't tell the boundary between like me and the AI systems I work with now because it's not like these clean handoffs, like your mental model merges with these. And I think that's that's increasingly what it'll look like. Circling back to where we started , I think what's really interesting also in this kind of freakout moment we find ourselves where everybody's like trying to use these things for everything and figure out what works and what doesn't and all that stuff. You think about organizations , I think about mental atrophy and also grit . Like as a journalist , I spent years and years stressing out and learning the hard way and all of that stuff . If one you have this new layer of kind of digital knowledge workers, which makes it harder for younger people to get hired . And two, the people who are getting hired don't have to, you know, bang their head against the wall all night to figure out a problem because they can just ask an agent to do it . How do you think about how this evolves in terms of like again, talent, especially like the very beginning of that funnel, the twenty two, twenty three, twenty four, twenty five kind of fresh out of law school people . So I would say the biggest is if I think about how I learned to program like fifteen years ago and I was learning to program now with these tools , I would be a thousand x better engineer than I am now. And so I think these tools themselves are actually the single best way to learn how to use this technique. Like I actually had someone I was at a conference on Thursday and they asked me like, do you have any advice on how to use these tools or what resources do you need? And I was just like just ask them. Like they're so good that they can teach you how to do this. And then I think sometimes people feel like if I'm using these tools to do this work, then I'm not learning it. Like my argument would be if you think of like the role that I'm doing now , I don't do most of the work . A lot of what I'm doing is how do I communicate to a lot of people how to do the work , but I still need to be accountable for it. And it's like, if you look at like as you get more senior in your career, you're increasingly doing that and it gets harder. So it's like what I think you want to think about is how do you get out of the range of things the models can do and start solving the problems that the models can't do, right? And for I think a very long time comp. Lanikeies are not going to be run by agents, right? Like you are going to want humans accountable for these outcomes. And so how do you learn this skill set of you're accountable for these outcomes, but you know how to delegate all this work? And I can tell you that skill set is way harder than solving a lot of these individual problems. Like I still struggle with that and I'm terrible at delegating. So I would say like if you're in a company, it's like how do you elevate yourself using these systems. And then I would say that to me like it's never been easier to start a company, right? Like before you had to convince a bunch of people to come work for you or you had to learn to program, it's like now these models are so good that it's like I think of like how much time I spent learning to program like product design, all these things so I could build the first version of our product. It's like that took me years and it was really hard to do ten years ago. It's like you could, do all everything I did in better in like thirty minutes now. And so I think there is just this huge opportunity to figure out how to use these models to do new type of work, start companies, or work in new ways that I'm actually very optimistic that if you use these models , like there's never a better time to learn kind of any of these domains and more importantly learn the new jobs that aren't going to that don't exist right now that I think will exist as these models get better . It was fascinating. It sort of really goes to the heart of the future of work , I think, you know , beyond the law . And it was really interesting how he kept talking about the models as tools and as add ons to the lawyers, which is obviously completely different to the story that emergency tells of Bots running around, you know, with their own personality doing their own things. It's far more my experience of using agents actually . It's the gap between the science fiction vision that is being spun out of here and the reality that you're finding with Norman or they're finding in these law firms , right? It's just like, oh, these agents could do everything. Oh my god, we don't need to hire any more humans . And just in the time in the paper this past weekend we had a story about a new grad whose offer was rescinded for their first job out of law school or at a new firm due specifically to AI. And I think that's becoming more common, isn't it? Yeah. I mean, you know, there's definitely an AI impact on younger people. You know, you can do more with less now. I think that's just true . But the question is like, okay, well if you have a more slack in the system and that meant whether that's money or time, how do you use it? Do you just use it to cut back and do and do the same thing you're doing before with fewer people? Or do you try to be more ambitious and like be like, well, now you're freed from all this drudgery. Now go do this really other hard or interesting or ambitious thing. But what's so interesting about this moment is like, you know, we've been talking about companies from Cloudflare to Meta , Meta has been a huge story out here because not only did they lay off eight thousand people , they reassigned another seven thousand for new kind of AI focused jobs and and they rolled out this new software , effectively spyware that tracks what their employees do, like their mouse clicks, like everything that they are doing on their computer to train AI models to be able to do those tasks , right? And that feels deeply dystopic . And in response , Meta has put out a statement they said if we're building agents to help people complete every day tasks using computers, our models need real examples of how people actually use them . So in other words, you know, it's fine. Don't worry, everybody. We're not going to use this data create bots to replace you. Yay . Nothing to see here. Nothing to see here. Correct. Exactly. But again, it's like, okay, you can do all that, but actually does that mean that you're just going to have a whole mill ions of agents running your company with a few humans around the edges making the big decisions? Or is this just all fantasy? The last thing I would say on this is because I did a piece based on this interview this past weekend like and subscribe at the time . Thank you . Thank you . But I talked to one of the main professors at UC Berkeley Law. And they're one of the top twenty firms in the country , they just rolled out a new policy that takes effect now . You can't use AI anymore . During your education, he said you can use it for specific purposes like help,ing you prepare for class and stuff, but he's like basically the thinking part of law school is the thing you need to learn , right? Like and my wife says this all the time, what law school does is teach you how to be a lawyer, how to think like a lawy er, how to analyze a problem, do a bunch of reading, find the holes in the argument, and then create a counterargument, and then you go forth . And he was like, we started out with a really permissive policy three years ago and we just saw like the quality of the work we were getting back from people was really starting to suffer . You had students just like citing non existent cases and what started as a grammar check on a paper became an entire AI rewrite that can completely change the voice and was just not genuine and he just like basically the thing they're worried about is what anthropological is called cognitive atrophy . Like you're just not thinking, right? Your brain turns to jelly and forget how to use it. Yeah, and especially for law for the law, you need to have that like kind of legal mind set that kind of comes through three years of, you know, suffering through university and trying to figure that out. But they got a lot of like, you know , they announced this and a lot of the big firms called them be like, what are you doing? We need these kids like, you know, converse and they're like, We have AI courses. Like these are these kids are going to be AI pilled, but when we're talking about the training of how to be a lawyer , you can't rely on AI for that. So it's a really it',s actually a really interesting nuance because they're like, we're we're in Silicon Valley. These kids have to be very, very proficient with all things AI and we are preparing them for that. But the actual like almost like the mindset that we instill in them has to come the old fashioned way. But before we go, before we go, one just and this has kind of popped out while we've been recording or this on Monday morning , anthropic, they have filed to go public. an email that they sent to everyone, I can only imagine because we've both got it . And they've put a blog up saying that it has conf identially submitted a draft S one to the SEC. Not that confidential anymore. Guys, this is secret. Don't tell anyone . Read the secret press release. And says you can't read my own. Andrew confidentially submits draft F one to S one to the SEC. No numbers on it. Yep. So we don't know anything, really. No numbers and no sense of when it says, you know , we'll sort of wait and see depending on market conditions and other factors . But there you go. And they were just valued at nine hundred sixty five billion , which critically is more than open AI was last valued at. They have leapfrogged open AI at least for now . And apparently, according to the reporting I've seen, they've gone their annual kind of run rate of revenue is now forty seven billion dollars , which is insane because it was like ten billion like six or nine months ago or something. So it's just like this company is growing so fast . But again, I think this links back to this whole conversation we've had this for the pod . It'll be really interesting because I think you're starting to see companies being like, What am I actually spending my money on? What is my return on investment ? And you're starting to see companies like Uber, there's I don't know if you saw this, their CE COO said we blew through our entire AI budget in four months because all our engineers are completely AI pilled. They're using these clawed code for everything . They spent through hundreds of millions of dollars and he's like, not sure it's worth it . He's like, we're not seeing like, yeah, we're shipping more stuff , but like we're not seeing this kind of come through and like, oh , that means we're getting more people riding in our cars or we're making more money. It's really interesting that they're going to try to go public at a time when these questions are starting to pile up of like these bills are not insignificant . And then for the financial world, everyone's going to be looking at the numbers. It'll be really the first time we'll have had a proper sense of how the company is doing beyond what they've chosen to tell us. Exactly, exactly. So fascinating . Well, I think that is it for this week's episode of the Chinese Tech podcast. If you are enjoying the show, drop us a line and let us know at techpod at the times.co dot uk that is techpod at the times dot co dot UK Yeah, tell us what you thought of today's discussion and if there's anything you'd like us to get into, drop us a line. We will see you next week. See you next week. Bye bye