OpenAI and Meta E9: The Career Bet Every Engineer Must Make

Philip, in your article, AI killed the IC. Your thesis is that traditional roles are dead. Your U-Rote. We have passed the event horizon. Pandora's box has opened. The genie is out. There's clearly no going back. Are we headed towards a system collapse? I debate with technical friends all the time. Whether or not this will replace Johns or whether it will augment Johns. But I'll posit this. What is certain is Johns and Software are about to change at a per-anim basis faster than it ever has. We're about to enter a world in which half the people will have an amazing time. There's going to be another half the world for which the coming world will be really painful. They will hate the way the world has changed. Those people are about to have a very hard time. The one call for sure that is wrong right now is that your job will stay the same. Philip, thanks for coming back. In your article, AI killed the IC. And everybody's going to become a manager. U-Rote, we have passed the event horizon. The genie is out. Have you had your coding agent moment? Yeah, I mean, I've been using AI coding stuff for probably a little over two years, starting with GitHub Co-Pilot. And back in the day, it was tab completion. E-NOMs and sort of like-- and it seemed to me like it saved a little bit of time. But I had a tough time convincing many former coworkers that AI was going to play a significant part in their code. But I think as you've seen in the last even two months, there's been a radical awakening. I think more people who were skeptics traditionally have now opened their minds to the possibility that this thing could be more than a passing fad and that it could actually do significant software work instead of just completing little sentences and stuff like this. Just so that we don't date ourselves. Today is January 29th, 2026. It takes a couple of weeks to produce one of these things. And so if we are already with the pace of everything, if we are already saying some out-of-date things, that's because of the date. Today, I feel like I need to put that caveat in there. This was recorded three months before the singularity. [LAUGHS] Oh, wow. I mean, this is going to age like milk, I hope. You're noting that everybody software developers are turning into managers. And I wanted to get your take on this specifically because you were in E9 at meta, and then you famously took a demotion to E7 so that you could become an IC and code again. And so you went the other way. I think several times, or a couple of times at least, is that the idea that you will be an IC? Is that a dead concept now? I feel like in my career, I've switched between managing and being an individual contributor probably six times. I do think the traditional role of the individual contributor is probably over. And that's the role of you're focused in your office, you're solving a technical problem, you're talking to two co-workers about trade-offs, right? Because if you think about the time you spend with AI nowadays, a lot of that time is what traditionally managers do. You're prioritizing between different agents. You're trying to decide between two LLMs that disagree on the right approach for something. You're sort of planning out the next week. You're essentially delegating almost all of the work. So I think the work has now become metalwork in the same way that most managers jobs are metalwork. Is it that ICs are going to become managers and managers will no longer be needed? Or is it that managers now don't need ICs? I think that you're going to find both. I mean, we've all had the experience of walking into a non-technical business. Let's say plumbing or HVAC or dentistry. All software developers have had the experience of walking into such a business. And being like, man, if you gave me one week with your business, I would have radically improved everything you're doing. Well, that's our hubris. We think we could, but then when we actually get in there, it's just like, oh, there's some use cases. I was not thinking about that. Yeah, like the problems are hard, like you say. But I do think on one hand, software engineering will grow into those domains in ways such that dentists can then do their own work. Like I heard a great episode of, I think, the hard fork podcast in the last week or two, where they ask non-coters to send in their stories of AI success. And one of the guys said that he's a plumber, and he wrote himself a customer scheduling piece of software with AI only. So those sorts of people back to your question about will managers still be needed? I think when we expand into traditionally non-software domains, those will be new business that is additive. I think we'll still need managers in the sense that even today organizations with managers have managers of managers, and it's like N levels deep. I think that's a human coordination problem that will still need probably humans for a while to solve. I wonder if we are on a pit stop, or we've reached a destination with the idea that STEs will become managers. I think the naive thing is like, why can't the AI be the manager? I don't think that's naive at all. Like I've said the same things over a number of years, but sound less and less crazy over time. Which is what I've told people, and you and I discussed this in the last time we chatted is like, will this thing never be able to do this stuff in 3,000 years? Like by the year 3,000? Well, it just never. Are there fundamental things the AI cannot do? And there are very, very smart people who feel that yes, there are fundamental things. Like Jan McCune would say. Like large language models will fundamentally not be able to do these sorts of things. So there are people who say that the LLMs will never be able to do certain things. But I think there are two problems with that argument, even if it is true. One is that there are definitely people that AI will not be able to exceed for a long time. But most of the world is not that type of people. Take back when Deep Blue and Casparov, right? You could say, oh, yeah, well, AI still can't play better chess than humans, right? But the fact is even in the Deep Blue days, Deep Blue could beat most humans at chess. We're not asking the question, will one human being be able to exceed some aspect of AI for the next 50 years? Well, there exists one human who can. I'm sure in every domain there will be many humans who can. But if it still is better than 90% of the people in any one domain, 90% of managers, let's say. If you have a manager that's like Dilbert's manager, can AI replace that probably, right? In the past month, I think I've realized that I have maybe been a Luddite and denying AI coding and AI agents. And I realized that I've been moving the goalposts. Oh, it's not going to be able to do leak code questions. It's not going to be able to do math, limpia, questions. It's not going to be able to refine requirements. It's in all of yada, yada, yada. And I realized that there are two types of limitations. There are temporary limitations and there are fundamental limitations. Every time I move the goalposts, I just latched on to the next temporary limitation. Like these arguments that people would say, they'll be like, yeah, the AI is the worst it'll ever be, because it's just going to get better. It's literally monotonically increasing to you. How can you say that it's not going to be able to replace? And so then I start searching for fundamental limitations. Well, I guess there are two types. There are fundamental limitations and there are temporary limitations. But the temporary limitation is one where I'm going to die before it actually gets there. So you talk about the year 3000. We don't need to go to the year 3000. I'm not making it to 2100 more than likely, right? Or if I do, it'll be just a tiny little bit. So I just need to outrace it for my working career, like my children and my children's children. That's a different story. If we take a look at fundamental limitations, I have a good friend from college. He's a radiologist. And for 20 years, I've been telling him, your job is going away because you have rich training data. You literally have everybody's MRI or CT scan or X-ray, right, the thing that he looks at. You have a rich set of diagnoses. And then you literally have their training. The training is codified. It's written down. It's part of his training. And so why is it that he should have a job in this situation? It's a lot like the coding thing. The thing with code is you express yourself with language. You can go and scrape GitHub. There's a lot of priors. But we have seen that radiologists have not yet been replaced. I think Carpathi said, actually, their jobs are safe for a long time. And when I thought about it for a moment, I think what it came down to was there is an accountability aspect associated with the job. He has to put his good name on the MRI that he's reading. And if he says it's cancer and it's not, or vice versa, that's on him. Yes, he can go and get insurance or whatever. And so I think accountability is a thing that, by definition, you can't have an AI suffer consequences. And so, yeah, you can be like, hey, here's my bank account. The prompt is go and build a business for me and make me money while I'm watching the game. But if it doesn't do a good job, you can't fire that AI. Yeah, I do think you have great points. One is regulatory, right? Like the reason a radiologist is still required is partly regulatory or in other industries union, there's a demand like the way that the TFL in London, even for the radio control, the remote control, the tube trains, they insist not only on one human driver in it, that does nothing, but that human driver in order to be kept awake and alert needs a companion to ride with them as well. So I do think you're right. That regulatory and union-like behavior will happen. But I also think when it comes to the personhood of corporations as an example, that's a humanity a long time to even consider that seriously. Corporations have that same problem as can they act as individual actors, can they bind themselves? to contracts and whatnot. And once we agreed on allowing the weird concept of personhood of corporations, a lot of innovation was unlocked because of that. I think we might see a same thing in AI. Is at some point we might decide on at least legal personhood for some aspect of them, to bring that accountability back to the fore and to have legal structures around it. I mean, what does it mean to have a model take on personhood? Because I mean, just think about the architecture right now, your API token is against a model that's housed in some data center. What does it mean to suffer consequences? Where is open AI? And what consequences is open AI the model going to have? Yeah, great question. I don't know how to split that. Like for instance, if your Tesla starts being able to rent its own self out as a robot taxi and it ends up accidentally hurting someone, like is that your legal fault for letting it out? Is it Tesla the corporations fault? I don't know how we'll split that. But I do think that people will increasingly use agents in a way that will bring this legal question to the fore because someone will use Claudebot eminently to do something that ends up being disastrous, okay? Like, and this will go to courts at some level. I'm not sure what the answer is, but I am sure it will be tested. So I spent 18 years in Amazon and I know what it's like to use an issue tractor that feels like it's fighting you the whole time. They're slow, they're clunky, and eventually people just stop keeping them updated, which means that your data goes stale. That matters way more now than it used to. Whether it's engineers using cursor to write code or PMs using chat GPT to draft specs, they're all pulling context in from somewhere. And if that somewhere is a system people avoid using, the output suffers. This is where today's sponsor comes in. Linear is purpose built for product development. It's fast, it's simple, and it fits how people actually want to work. Use switching from Jira or GitHub issues, end up logging up to three times more work, just because it's no longer a chore to do it. I use Linear to run this podcast and to create all of my video content now, and the difference is night and day. It took me about 60 seconds to set up, and I didn't have to train my team on how to use it. This has moved the conversation about the million projects that are going on away from Slack, which has been a game changer. If you've been thinking about making the switch, check them out at linear.app/ale. I think of airline pilots today. They have autopilot. The autopilot is pretty good. We still have two people in the cockpit. And I don't think that that's a union thing, right? I think that's a, like, we're talking about human lives. And so I would like two people in there. If one of them goes to the bathroom, then the other person, you know, somebody needs to thank our, like, I think this is just, this is like, good procedure. We didn't need the LLM reckoning to develop that. And so I think that this high stakes, but with humans in the loop, it's unclear to me whether that's a fundamental limitation or a temporary limitation. I'm glad you brought that up because I haven't thought about this until you brought it up. Like I feel the same as you do about airline pilots. I would be very hesitant to get into a plane that had no human pilots. And at the same time, if you were to ask me, even for a cheaper ticket, I told you. The first ride's free. Even if you were to ask the last one, but consider this, if you were to ask me today, driving home from this interview, would I rather the road be a hundred percent autonomous vehicles except for me, or would I rather drive home with all the human drivers that I'm going to encounter on the way home, on the highway? I would much rather be on a highway with only autonomous drivers out there. And me, of course. But what I mean is what is the difference that causes me to feel that comfort around cars, but not around planes? I think it's because I don't know how to fly a plane, but I do know how to drive a car. I see. It seems mysterious enough for me to rely on the expert knowledge of someone else who's been trained to do it. Yes. Yeah. I think you're right. Coupled with, I would say, I know how bad drivers can be. We demonstrably don't have pilots that bad, right? But you know, drunk drivers, texting drivers, like all of these things, we know they exist out there. And radiologists fall more like you say on the pilot side, which is there aren't incompetent ones just operating all the time. But how much of human employment is that sort of employment versus how much of human employment is literally Uber drivers driving taxis? You know, I think about, you know, the use of AI in war, right? Yes, I would like to win the war or the battle or whatever the objective is. But if it goes and there's a bunch of collateral damage based off of like, AI's, you know, it's just going off the rails or it made a calculation that, you know, the objective was more likely to succeed with its actions. The AI can't be the one that takes the consequence for that. It needs to be the person that gave the AI the control. We have a psychological barrier against accepting machine fault. We can talk until we're blue in the face about how much safer per mile Waymo is than the average human driver. But the moment that it hurts someone, we want to see blood, right? I think this is a psychological thing. The numbers don't appeal to us when people blab at us about how safe these things are. If a human wrecks something because he was drinking before he drove, we have like, if you can believe it, some semblance of forgiveness for that person more than we would in AI that's not drunk. And I don't know what about it psychologically? Maybe our ability to empathize with or imagine ourselves in that same condition in a way where we cannot imagine ourselves as an AI making mistakes. I am totally like with the driving argument, I get it. We can look at the numbers collectively and say it's an order of magnitude safer. And that's kind of what I need before I relinquish control. I get that. But let's talk about law and life and death, right? Like, we still have judges. And what is a judge other than like a human making a decision on behalf of society? Nobody's talking about like replacing judges with AI. But consider this, you know the famous study they found out where judges in America were giving harsher sentences before lunch than after lunch. This is like a well known phenomenon, right? We accept that weakness in our human judges because we sort of trust their, you know, zen special human qualities for making decisions. But I don't know if I were brought to a court case where a judge were to reside on the case. I don't know if I would actually prefer something that I feel is impartial. When I talked to Uber drivers early in London. So when I went to London, Uber had only 4,000 cars there. And black cabs greatly outnumbered Uber's by 10 to 1 back then. But I lived in London for five years and by the end, Uber's were matching in number with black cabs. I asked some cab drivers who used to drive for many cab companies like why are you driving for Uber now and not your mini cab company. And one of the top responses was the problem with driving for a mini cab company is the dispatcher who is human will give better jobs to the people they like and worse jobs to the people they don't. One thing that many drivers for Uber liked in London was the sense that an objective thing would assign them roles, right? In the same way that you see these studies that young people now would prefer an AI therapist over a human therapist because they don't like feel the eyes of judgment on them, right? So I do think there's a sense in which you're absolutely right that we love so many things about handmade things, human made things. But I do think there are many domains in which we would rather have the objectivity of something non-human. I'll push back on this. When I was in London, so I used to go quite a while. I didn't live there. I figured out how to hail a cab. You put your hand out like this and then they drive up and then they're asked, you don't go in like you would in New York or something like that. You're like, "Hey, I'm going here." And they'll be like, "Oh, this place, they confirm they're like hop on in. Then they will unlock and you get in." But they famously have this concept of the knowledge. The knowledge is this 10 radius thing around trying cross, I think, where they're expected to know every single street, every single nook and cranny of the city. It takes them years to do it. There's this whole journeyman sort of thing that's associated with it. And I was like, "You know, I was like, dude, Uber is going to eat these guys' lunch." Right? But then I hopped into an Uber, maybe this was probably 10 years ago, and Uber had no idea how to navigate the traffic. There's the River Tim's and it just, it's in this weird sort of like geometry within and geography within the city. And there's multiple ways to cross it and to minimize the driving time does not imply that you minimize the distance driven. And so I trust the guy with the knowledge much more than I did the Uber, at least in that instance, because it was like a 45 minute cab ride when it should not have been. Because he didn't know anything about where he was driving. I'm trying not to be this like AI denier anymore, but I just feel like where are the fundamental things that like are safe and that we can't hand off to AI? And what are the things that are, you know, just temporary? It's just going to get better over time. Two thoughts. One is relative to the knowledge living in London five years. I had the opposite experience as you, but for two reasons. One, I've gotten into multiple black cabs where the cab driver with the knowledge has Google Maps open. Yeah. Like part one. Part two. Can you blame the guy? Right. Part two, even if you knew the map of London like the back of your hand, you cannot possibly have real time traffic information that would inform you about dynamic changes. So I think there's a difficulty there to overcome. But to your point of like this AI sort of denial and stuff like this, or maybe that's not the right term, I think that there is one way in which we who are living sort of middle class and above American lives do not want our doctor to be replaced by some AI chatbot that's hallucinating a third of the time. But I think at the margins what we forget is so many people in the world who cannot access that normally, giving them something that is right two thirds of the time is better than nothing. Right? I do think at the margins, there's also adoption that's possible. That is not going to be better than what you and I can access today, but is better for much of the world. Education falls in this category. Like, would it be dystopian to have my children educated by AI only? I would say yes, that feels dystopian to me right now. But for people in many parts of the world, for which getting an education at all is already very difficult, would that be dystopian? I would say not. Like, that's a great boon to them. So I think at the margins is where the improvements will come at first. Let's get back to software development. We'll come back to broader societal implications. Like, when did the switch flip for you in terms of AI coding and agents? I was probably very early in adoption of AI. And I don't know if it's because of laziness. I didn't want to type the enums or whatever. Or just my tech forward sort of like, "Oh, I'll give it a try," right? So I was relatively early. But I would say in the last year, I would say there has been a substantial change from me doing coding some of the times. To, I'd say in the last six months, I have written very few lines of code, right? So that was one change. In the last two months, I would say, with things like Opus 45, with the latest Clawed Code, with Codex 5. Two, I would say in the last two months, it has changed significantly from that, where I now trust the AI a lot more to do the analysis and the right thing when it comes to the code. Like, I'm not fully at the point yet of that meme where the guy's just waving security people through. So I'm not fully there yet. But I have definitely shifted to a point where the coupling of the AI smartness, along with two different code review bots on GitHub also chiming in with problems, I am now reviewing code even less carefully than I used to. So Anish from Andrews and Horowitz, so general partner there, he says that with AI, we are getting into this YouTube moment. So if you think about YouTube and its development, you know, circa 2006, 2005, everybody's like, why are you letting regular people make videos? What is stupid idea? Like TV shows and movies should be made by the professionals. But it turns out we were short on videos. Just look at your YouTube feed, like just look at how it's displaced a lot of the viewership for TV. He makes the same argument for software. Now non-tech people can go to AI and just be like, hey, can you just make this cool app idea that I had in my head? And we're very quickly getting to the point where like that's happening. One thing about this YouTube moment though is TV shows and movies didn't go away, right? You still have movie stars. But now we have this other class of people as well. And so I do think that when it comes to corporations, there's not a lot of profits or like things to get more efficient at when you talk about like the core competency of a particular service. So think about S3 or Azure. Do you really want to vibe code and not review? Like the core infrastructure that runs the internet? That wouldn't make a ton of sense. Yeah, I think you're right at the moment. But I'll tell you one experience I've had with code review bots over the last six months that I've used them. I think when they first started, when I worked at OpenAI, it has tried many times to do code review bots. When I first started at OpenAI probably two and a half years ago, the code review bot was so noisy and so bad. Like the false positives, the types of claims were nuts, okay? These days, if you have one or two code review bots running on GitHub, you will find that a lot of the things it brings up are things that you, as a human reviewer, genuinely did not think about. Once the quality threshold crosses to where you readily acknowledge it is doing some things better than you, then I think eventually we, like with automated driving, will flip over to only using cloud services that have guaranteed that no incompetent human has touched the code. You know, we do the opposite. Yes, we will, one day get to a point where we're like, what? You actually have humans? I'm not using your cloud service. Like that's nuts. Whoa. Okay, that's wild. I even thought about that. That the idea that humans touching the code is actually decrease in quality. I think we're going to tip to a point where like you will hide the fact that you have humans that intervene in your code. That you would be embarrassing. Yes. Yeah. Because the errors, that's wild. Yes. So, I mean, in the same vein of like temporary versus fundamental limitations, yeah, Boris from cloud code, he's basically saying, I don't write code anymore. Like right now I'm reviewing, but also I also have them do a first pass of review. Right. And so now I'm just looking at the outputs there. So I think that the current goalpost moving, now that I'm not in that group is humans will never be replaced because of taste. We all have this taste and this experience and it's, you know, we can't compress experience. Is taste a temporary or fundamental limitation of AI? I think it is temporary. There will always be some people who have genuinely great ideas, right? But every artist thinks there's a snowflake that there's snowflake, that their inspiration comes from the sky down, right? Like they get pissed when they see that the AI has looked at other art and produces what they call derivative art because they feel like their thing is not derivative. But how many artists like every artist looks at other art, right? That's how you gain that knowledge except because we can't attribute your painting back to the training data you use to generate your art, you think you're genuinely creative. I think taste is a little bit like that, which is like, if you consider what percentage of the co-workers you worked within the past, do you think have good taste? I think co-workers are great taste. I know that. I know that. Other than you and me, it's obvious that the world is full of people without taste. I would say that there are, of course, truly exceptional people, but that the average person working in the industry is honestly not doing that unique snowflake of a job. Like the average person, I think, easily replaceable. Now let's talk about this like this attribution, right? And this tastes like where did it come from? Where did it originate from? Yes, I think it's totally okay to be inspired by other people. That's just how art works. There's no original ideas. I think that's fine. But when you have AI reproducing the art style of a single person, and it's just very clear that you just lifted it off of the training data because it was on the internet, I'm not sure like I'm okay with it. Like think about the US Constitution. There's a section in there around patents. It says, hey, if you came up with a really amazing idea, the government's going to protect that idea for 10 years. And that means that you have rights over this idea. You can license this out to other people. But this way, we are allowing for innovation. You have a right to profit from your original ideas. And we have this whole arbitration system with the patent office, you know, where they're basically like, is that your idea, or is that actually from somebody else's idea and all these rules? That's a systematic thing, so that innovation doesn't go away. And so I worry that without legislation or protections, this is literally in a legal document from the end of the 1700s that they have this idea of patents. We're in 2026. And we have nothing when it comes to AI. Are we headed towards a system collapse because of the copying and mimicry that's okay? I think you're right that our legal system is probably not prepared for how to properly incentivize continued creation. Right? I don't know what the answer is, but I just think you're right. Like that is the case, right? You could also argue like patents, what did they last like 17 years or what have you, right? The number of years a patent last has not shifted a lot, but you could ask yourself the business environment and the iteration speed of business has obviously shrunk over time, right? Should patents still be 17 years? Like does that even make sense? Right? 17 years ago where was TikTok? Right? And so part of me also wonders like whether the system hasn't adapted the other way which is to accelerate innovation by shortening those timelines. I don't know what the answer is, but I think you're right that there is that problem. And of course the fundamental question of like how many things coming from a human? Let's say a human musician. How many of those things are ultimately derivative versus quote-unquote genuinely new ideas? I don't know what the answer is, but the answer I don't think is 90% of it is creative, and then 10% of it is derivative. I would guess that a lot of things are clever remixing of things from different domains or from different areas that are just a clever recombination of things that we'd like. I mean, if you want to talk about music, yeah. Like there's that thing where it's like CD and G, those chords on a guitar, that's like 90% of all. Four chords and the truth as it were. You're right. And so like yes, but I think the thing that we're grappling with is when the cost of imitation goes to zero, what does it mean to not have protections for the person that came up with some useful combination of things, whether that's notes or lines of code, there is no protection. And so like why would you release it to the world only to get sucked up by open AI or anthropic as they're crawling the web? I don't know the answer, but I'll pose a few things. There are many musicians for whom the music just has to come out from them, whether they make money or not. I think there's a type of person for whom creatively, let's say a software developer. There's a type of software developer for whom, if you didn't pay them to write the software, they would have to produce it anyway and it would be open source and they don't know how to make money from it, but it just has to be there, right? So there is a level of I think human initiative, I think in many people, that is unstoppable. Now, do I think they should be rewarded for their creativity? Absolutely. Like if we can create the legal structures by which those sorts of contributions are contributed, that would be even better. But I think that if everything goes to zero, which I don't think it well, but if everything goes to zero, I don't think that means all of us just Netflix and SIP big sodas and the Wally future is here. Like I do think one tenth of us will still be out. building things because we find them interesting even though we can't make a profit from them. I don't know if we are going to a world where there's no market. The dynamo that is a well-regulated market that is brought upon this wonderful era of humanity, I'm not a post-capitalist guy, I think capitalism is actually really good, we do need to add some rules to it so that power doesn't concentrate. It seems to me that markets will just move to the frontier. Software used to be very expensive and now it's going to be cheap. There was always this joke when I was coming up which is like, "Oh, your uncle hears that you're a software developer." During a family gathering, they'll come over to you and be like, "Hey, I've got this wonderful idea for an app. You coded and we'll split the profits 50/50." You're just like, "Do you know how much software costs? No, your idea is not novel or whatever." Now that drunk uncle can just go and do it on his own. Totally. Whatever that ludicrous idea is, what happens is you never have to buy another app or is it that there's actually going to be some skill to make a good app. If we go back to this YouTube example, now it's so much easier to create a video. I didn't know how to make a video six years ago. Now you can just go and do it. I can see somebody else's video, but I can't reproduce it because I don't have enough skill. My taste is good, but my skill isn't quite there yet. But we're rapidly approaching the situation where it's like, "Oh, actually, I can just imitate you and I can do that for free." I can extend it as well. Who's going to pay money for this app? It's unclear to me whether the market will shift and then the thing that is valuable is the thing that the AI can't do. Then it's just this forever-moving frontier. I think that there's a balance between profit and consumer surplus. So when the cost of producing useful software goes near zero, then I think it would be a malfunctioning market if you could make a really huge living producing software. Because that is collecting some form of rent. I think that you're right. That as the cost goes to near zero, that it will become all consumer surplus. Like people will get to use this stuff for near zero and they will benefit. People just forget the consumer surplus side of things, right? Everybody thinks about like, "How can I make a living?" Nobody thinks about like, "My ability to make a living depends on the market, not being so free and so full of competition that my ability to make a profit goes to zero." But when my profit goes to zero, that's the best consumer surplus. I'll give a concrete example, like 60 years ago, most American corporations in any large building had at least one floor of secretaries whose full-time job was to type memos, right? Eventually, word processors let everybody type their own stuff. Now, on one hand, I bet you 90% of those secretaries laugh at the poor typing skills of the people who replace them, right? But the thing is that it democratized the ability to type so that now more documents than ever have been typed by less competent people. Does that make sense? I think more software will be written by less software competent people than ever before. And overall, in net, that will be good for the world. Even though the people who are great at software will be the type of people who get together at coffee shops and complain how like the modern software is all crap because the way they used to build it is great. Just like these typists, probably get together and talk about typing 180 words a minute and look at all these crappy people typing the crappy stuff. I mean, the concept of mail merge eliminated that floor of people whose job it was was to take the forum letter and then type the address for Philip. And then now I'm going to type the address for Steve or whatever, right? So I get that part. I just, you know, you just blew my mind and you're saying like, hey, the idea that a human will touch software is going to be a detractor to that software. So it doesn't matter if you're incompetent. Everybody's competent now. And so software is like, is there a mode? There's no mode anymore. Again, the goalposts, movers are like, oh no, it's going to be taste. But then I think taste is imitatable as well. Yes, I think absolutely so. And you've highlighted a key question in my mind is all of these DJs who don't produce their own music, but instead preface the music they played from someone else with another one. Whatever. Like all the people who are claiming taste as their value add. Clearly, the world agrees. Like, yes, they have taste. But over time, like you say, will that be a domain that only humans can do for some reason? I think eventually, even if AI lacks that sort of taste, musical taste, let's say, remix the right things, AI is infinitely patient to mix 10,000 of those things and just select the two that's actually succeeded. So it will overwhelm you in sheer number and volume with lack of taste. When it comes to this AI stuff, one day I think I am so powerful. I've never been as powerful as I've been in this moment because of AI. And then the next day, I'm like, I am going to be replaced. And so my value is reaching zero. And I flip flop between the two like pretty much every day. And so take that on the micro level and move it to the macro. Like, where is our economy going? Is our economy going to increase GDP by 10 points? Or is the economy going to die? And this concept of capitalism and this economy that we've created, is that going away? Whether or not this will replace jobs or whether it will augment jobs. And of course, the likely answer is not 100% of one or the other. But I'll pause at this. No matter whether you believe it will replace jobs or augment them, what is certain is jobs in software are about to change at a per-anum basis faster than it ever has and at a comically faster rate. So regardless of which side you believe, if you think your job is going to stay the same by this time next year, you are almost definitely wrong. And if you aren't taking steps to either bet on one future or the other, the augment future or the replace future, right? If you aren't taking active steps to bet on one or the other, it is almost certain that you will be left in the dust. Okay, so you're saying if you're on this like AI is not going to change your job, that's the only, that's the wrong move. That is the only for sure wrong move. That is guaranteed wrong in the same way as the woman who proudly took out all her cash before the 1929 crash. And when the interviewer asked her, wow, that was really insightful. What have you done with a sense? This was decades later. She said, I still have that money. It's like, yeah, well, you made one good call, right? Again, we're end of January in 2026. I think everybody has sort of had their cloud code moment. Twitter at least or X in the world, they basically have been like, okay, the writing is on the wall. And there's this sort of existential conversation that's going on. And I like your framing. You're basically like, okay, well, I think everybody collectively needs to make a bet on Polymarket or a Calcium or whatever, but no, just with their career. Are you going to be in the augment camp, which means that, hey, we're not going to get replaced in our lifetime, or these are working life? There'll always be a human in the loop. Or are you going to put your bet on the replace future, which means that you will take a different set of steps, but even in the replace future, I think you're trying to understand like the second and third or consequences of replacement, and then maybe position yourself in a way to benefit from that. Totally. If you're in the augment camp, you should be doubling down on running eight agents at a time right now the way the best people are. You should be doubling down on custom subagents. You should be paying for the max version of cursor, cloud code, and two other things. You should be paying $900 a month because you are like the most awesome, right? If you are in the replace camp, you need to be actively changing your skills and capabilities, assuming your job is going away. And you know what I find is so funny, is these business inside of readers or these upper middle class like tech people, they're talking all the time about how we should encourage people to get back into the trades like plumbing and stuff, physical stuff, right? Not a single one of them is going into plumbing. Yeah, not a single one. It's a joke because plumbing is well, we like to tell other people this thing's coming for you. You should go learn welding. And it's like, okay, like if you truly believe that, you should go learn welding, right? And I think that's the difference is that many people are opining about this, but even people holding certain opinions are clearly behaviorally not acting like they actually believe that is what's happening. Yeah, I think that's true. I think words are cheap, right? Like look at their actions. Are they actually getting into the trades? Which by the way is definitely not going anywhere, right? But also like, no, they're not. I do think that if you are, you know, if you're a software developer or trying to get into the industry, if you are not embracing AI tools right now, you are losing. I think if you think about business trends the last 45 years, right? We went from a time in the 80s where ideas were everything. Like, oh, like brilliant entrepreneur has a neat idea. Then we went through a time starting the year 2000, where all the hipster CEOs and stuff, all of a sudden obsessed about execution. Execution is everything. Ideas are a dime a dozen. I think one thing that will resolve itself within the next two years is we will soon find out our ideas a dime a dozen or is execution like where it all matters. Because if we move into a world where execution is easy and all it takes is the taste to makers with the great ideas to produce things, we will pretty obviously see that the world will shift that way. But there's also a possibility we will find out that actually the world is not full of brilliant ideas. Yeah, right. That the world is mostly home, right? In which case, execution becomes the limiting factor. It's hard for me to think that if execution costs go to zero, that ideas are exhaustive. We spoke briefly about this joke space. The space of all jokes. Every joke that could possibly be told. I think AI is terrible at exploring that space right now. It's not like protein folding. It's not like this new spade of coding tools. I think when it comes to coding solutions, so coding solution space, we just need a solution that works. That's all we care about. Then you're like, "Well, actually, the solution to be all faster than you prompt it, and then it goes and finds the first solution that's faster within solution space." I think like poetry and jokes. I don't know. We all talk about, this is where they think the jury is still out. What's your take on when an AI can just like on demand make you just like laugh your sides off with a joke? I'll add one more thing to that, which is there's a difference between when I think AI will be funnier than Dave Chappelle versus when I think AI will be funnier than the average open mic night in Seattle. That is probably already funnier than that. You know what I mean? Yeah. The reason that's important is one is a question of extents. The other is a question of pragmatism, which is like, yes, maybe the AI won't get that funny, but it only needs to tip the thing. Like it doesn't need to be literally better than every human who could possibly do the same thing. It becomes economically transformative and workforce wise, completely disruptive. The moment it tips something. The moment it does something better than many people. Again, the jury is still out. Again, if I'm moving goalposts, that's fine. I don't see a world where it writes poetry that blows your mind because it's the big companies and they focus all of their attention on code. And just they just care about coding solutions. The fact that we trusted enough to like check code in, it is not done in an exhaustive search of code space, code solution space. It's done a search where I found one thing. And it decided that like, hey, I can take this problem and I can split it into subproblems. Again, using some sort of like naive search across that space at inference or test time. And then like they recurse on that smaller problem. And then you can sort of show that you can just stack all of those things together and then you have your product or website or whatever you wanted to build. But it doesn't work like that with jokes. Even with things like protein folding. Yes, we don't use LLMs for protein folding. Yes, we use artificial deep mind stuff. We're still using compute on it, right? But it's still even though the numbers are really, really large with protein folding, it's still bounded. Proteins can still only be folded in so many different ways. If I were to paint an optimist picture, what I would say is there's also a difference between would you buy something because it's better or would you buy it because it's human? So take like handmade rugs or take the whole big a row that happened like two decades ago when Starbucks replaced the barista packs coffee with the auto one button drain coffee, right? I bet on average that coffee is probably produced with higher consistency and skill. But yet we often am willing to pay more for a lesser product because a human being was involved. We might move into the opposite of what you're talking about before or where I just determined. Yeah. And so this is where I'm unclear. It's like I can't tell if that's a luxury belief, you know? Because the only type of people who buy hand woven carpets are the type of people who probably pay to get bridges disassembled when their yacht can't fit through, right? It's like they like the idea that a human had to labor to do something, right? Like will there come a day when you will buy digital art from a person for your triple A title, you will buy, you'll pay more like you do for fair trade coffee, right? Well, you pay more for a human to have produced that digital art than you would for AI and would you do it out of a sense of principle? I don't know. Like we could go that way, right? And so we could find ourselves in a world where humans are valued for their humanness only and that our commercial value no longer comes from our ability to beat any machine at something, but merely because we are trying to contribute back to this world of ours. I mean, I think at the end of the day it comes back to if we live in a capitalist six society still with markets, who's going to open their wallets and why? Those arts exists. Yes, there are some people that will do the art without the money, but like, I don't know, we're in a building right now where there's a ton of artists, you know, there's some glass blowing people, there's people who make toys over here, there's some painters and stuff like that. They need to pay the rent on this building. And so yes, they want to do, they want to express themselves and you know, they take pride in their craft and yes, but there's a reality of the world. Right. Okay, let's shift over a little bit. I just installed Cloudbot, now called Moltbot. I have not been this excited about AI technology since probably the first chat GPT, like 3.0 when it could drop. You know, I was one of those people that went and bought a Mac Mini and installed it. I first heard about Cloudbot on Friday and it's Thursday right now of the next week and I feel like I'm already behind. Like for some reason, when we talk about this AI stuff, like a week feels like a month, a month feels like a year. For the first time, like, it feels like I have a human. Like AI is moved away from tool that's integrated with my IDE or this chat interface, right? That's past the touring test, but really it's like, I don't care about that. Like, can you do work to like an employee? It really feels like, and I'm like texting my employee. So I think really there's going to be a reckoning. Why are software developers like the ones that need to like have this existential dread? I do see a day where if you really think about it, if you really want to reduce it, there is nothing special about a human using a mouse on a web browser to point and click and type. It is literally the same thing as those that floor full of people typing on typewriters to do the mail merge. What are your timelines for AI doing your taxes? Or AI replacing, just replacing, like, the need for anybody to do like regular office work? I don't honestly know because I think that a lot of it depends on a few things. One, you mentioned, which is the accountability of mistakes, right? If your cloud buy goes off and does something terrible, like, I guess right now the accountability is clearly yours. Like, you chose to run this, right? But you can imagine it getting more complicated if other people's stuff is involved. I think another has to do with whether you will do more with more or do the same with less. Take your business, right? With where you see this going, do you see yourself reducing staff in the next year or do you see yourself expanding the number of things you will do in the next year while keeping the same staff? That's a great question. You know, I stepped away from software about two years ago. I was like, cool, like, let's make this business and, you know, I'm going to make a bunch of YouTube videos and I'm going to start a podcast and I'm going to do all this stuff. I ended up working more than when I worked at Amazon and my corporate job grinding for somebody else's code. And I'm like, but then like, I'm getting some amount of success. Now I have to ask myself this question, like, why am I in this game? That's really top of mind. Do I want to keep it like a small lean team, which just like limits growth, but then like allows me to do like spend more time with my family or work in my golf game and all of the other like things that I want to do with my life. Or do I want to min max it and then just be like, how big can I take this thing? Do I want to be a Mr. Beasts or a Mark Rober or, you know, do I want to, you know, get into the software game again and like, build a startup where I'm like vibe coding everything. I think when it comes down to it, do I have to make a choice if I can just buy more Mac minis? And also I might throw in one other caution from the medical world, which is, you know how every medical innovation in the IT medical space always the press release says this is so doctors can now finally spend more time with the patient, right? Like medical note taking radiology, every one of these is like so that the doctor can spend more time with the patients. But objectively, what has happened as we adopted all these tools is the doctor spend less time than they ever have with patients. Okay. So everybody says I'm going to have more time with family because this tool is going to take me out of happy to do it. Actually, I guarantee you what will happen in your personal life is what happened with doctors is you swear to yourself, this is going to buy you freedom. It's not you are going to max it out, you know, unless you are like zen like in your control of your own life, you will go the way of medical doctors, which is you will tell yourself this tool will make my my life better because now I can finally focus on priorities important to me. But instead what it will do, it will empower you to chase the other thing even harder. That's super interesting. There's a salary subreddit where people post their salaries. It was this dermatologist and he's like, yeah, you know, we we grossed two million this year or something like that. And everybody's just like, I'm a doctor. I'm not making two million. How are you making two million? And he's like, well, I treat it like a business. You know, we do a bunch of elective stuff like Botox and you know, all of the skin treatments and stuff like that that are elective. And then he's like, you know, I invested in a scribe. And so the scribe allows me to do as many patients as I want to. So exactly what you're talking about. It's like they're not spending more time with patients. They're just taking on more patients and moving people through the system a little faster. The analogy I use is like smoking to get ahead either in corporate life or in business. You kind of have to do something that's not good for your health, right, which is to elevate your stress level. And then it's just like, okay, cool. There's this bridge or there's this horizon that if I just got to the horizon, then I'll stop, right? Whatever revenue target you have or whatever success means to you. But then like the expectation is that you're going to stop smoking all of a sudden. No, you're addicted to that stress that has been given to you, right? Like all your life has been min-maxing all of this stuff. And so, you know, I think it's right. I think, you know, do I want a lifestyle business or a growth business? That's, you know, typically how it's framed. I think I want a lifestyle business, but like, if I had all of this extra time, am I really going to do all of this optional stuff? Or am I going to notice that I can run this a little bit more efficiently? And then even if I try to artificially constrain the size of my business, like, I'm still going to want to go in and fix the window because the window is broken. Absolutely. You know, Tim Ferriss got famous for the four hour work week, right? But I guarantee you he works harder today and longer hours than he has ever worked. Okay, the man who made a fortune telling you how to spend four hours of your week so that you can spend 36 hours doing amazing life stuff. Okay, he himself, I'm sure works harder today than he ever has. So I think this core question comes down to like, the tools don't themselves, you know, Keen said like decades ago that he thought that by the 1980s, we'd all be working 15 hours per week because productivity gains were such that, right? His productivity gains estimates were spot on. By the 1980s, we had enough per capita productivity, exactly as he projected. But it turns out that as you know, in the last 40 years since the 80s, white collar workers work more hours now than they did in the 80s. Even though our per person productivity is way, way higher. I think what we're seeing is a battle between our appetites and our needs. And it turns out that our appetites are limitless. Well, maybe that's where it comes from, you know, I think. Everybody's like, oh, the good old days of the 40s and 50s, where one income could like buy a house and you could put food on the table and you got a car and stuff like that. It wasn't the good old days for everybody, first of all. But also, a house was like 1200 square feet. That's not enough for a family, for in today's standards. That's right. Yeah, you think about modern times in upper middle class America. Even having two children share a room is like, oh, I'm so sorry to hear your kids have to share a room. Like this expectation, where's my brother and I had a bunk bed, right? And like the whole thing, in 1950 to your point, the median new house build in America was 950 square feet. And not only is that a third of what the median house build was last year, right? But the average household size was more than double. Meaning it's far more than a third. It's like six acts, okay? It's way more than two kids, right? Absolutely. So like, I think you're absolutely right. Like, our expectations have ratcheted along with our appetites and desires and that apparently can consume far more than we are able to produce. So to the friends who debate me all the time, that is their point. About this thing is not going to replace anybody. It's going to augment everybody and our appetites for even more is just going to grow to such an extent where everybody will still be able to be fully employed like just satisfying these desires. Yeah, it's just like, yeah, if we go back to the replicator thing, it can make Earl great tea. But there's probably like, are you saying the replicator can literally make anything or is there a limitation to what it can make? The limitation for what it'll make is the thing that people will desire. Yeah, and also think about positional goods. People will always want positional goods. And that's just in our biology. What's a positional goods? Oh, sorry. I got the best seat for this basketball game. That's a positional goods. Only one of us can have it and no matter how much we pay for it, one of us has that, right? Those sorts of goods are independent of the cost of producing them. They are merely a matter of who else can bid for them. I think we're going to move more and more into worlds where positional goods are still, you know, our ability to want something simply because nobody else has it. Is the thing that, by definition, you know, is always going to push the frontier out. Yeah, no, that makes sense to me because like back to what you said around software, software is just it's simulating hardware, right? We have these like nan circuits or whatever. It's just like, oh, actually, you don't need to fab that. We can do it on the software layers. This not as efficient or whatever, but we don't actually have to make a board for everything. And so now that we're seeing software costs go to zero, eventually who knows how fast and what the side effects are going to be. There is going to be a thing that is still going to be valuable because we're describing it relationally. I just purchased tickets to the Super Bowl. It cost me an arm in a leg and hopefully the Seahawks win by the time you see this. But man, those things were freaking expensive, dude. It's crazy. The fact that software costs go to zero is not going to change the Super Bowl ticket prices. Absolutely. Yeah, and Super Bowl prices are a great example of that. If you look at Super Bowl prices four decades ago, right? Like, why are those things happening like this? I think Super Bowl prices are a reflection of increased inequality in our world because the wealthy now are competing for the same pool of goods. Like, the Super Bowl stadium is only so large. If you want to be there live, they're only so many people, no matter how wealthy your country is, right? And so that is an example of a type of good that will only spike in price. So we might be entering a world soon where some goods are going to be laughably expensive. And yet people will go by them. You know, you had brought up the example of the voice actor. I don't know if she was a family friend or you were related to her. But like, I just told my Claude bot, Molt bot, to go and make me a voice representation of me on 11 labs and it just went and did it. And then sent me a message back to me in my own voice. Like, I, you know, I, yes, I could have gone and done that on my own. But like, I just told it to do it and then just went and did it and then sent it back. Maybe the thing is just like, hey, this era where your job is like in your role is this set thing. You know, you can do this for the rest of your life. That era is just going. Like, I think people, you know, when I was a little bit younger, they're like, hey, it's not like your parents anymore. Like, you are likely to have like five different jobs or something like that over the course of your life. Maybe that number is the thing that's changing. Hey, like this idea of a static job with a static role, that's going away. There's this expanding frontier of things that are going to be valuable in the future because just the universe is finite. And so maybe the meta is like, hey, you're just going to be open to whatever it comes. Right? And like, that's just the world is just like, hey, you're going to have 50 jobs before you retire if you just graduate college in 10 years. I think so. You know that whole idea of maximizers versus satisficers, like one way to divide the world based on personality. You're going to have a hard time being a happy person if you're a maximizer in the modern world because there are too many choices. I do think when it comes to the big five personalities, one of the axes is something like openness to new experience. Right? We're about to enter a world in which half the people will have an amazing time because they on the big five have a very high openness to new experience. They're going to love that the world changes their job every four years. Okay? They will hate the way the world has changed because it will demand that they either change their jobs every four years or else not make a living. Right? It reminds me of a short story. I think it's by Ellen Lightman or something like that, which is, you know, he thinks about like if we were truly immortal, there are two types of people. One person that is like they will want to be a doctor. And then when they're done being a doctor, then they're going to be a dentist and then they're going to be a lawyer and then they're going to be an adventurer. They just sort of cycle through all of the things that they could be in their life and there's another set of people that are basically like I'm good. I think what we're saying here is like the former type of people are the ones that are going to be the best positioned in this new world of AI. They will thrive because our earth is going to undergo a Cambrian explosion. Right? Like what's going to happen is the environment, the work environment will have the equivalent of an asteroid hit. You are either the type of creature for which change is suddenly to your benefit because in that environment, you are faster to change. You evolve faster. You exploit new opportunities a lot or you're going to be the type of person who keeps trying to make the old thing work and you're going to find out very slowly that the old things not working again. I think about like madmen, it's like cool. They work in an ad agency. You show up to work in a suit. Sit down. You pour yourself a whiskey. You light a cigarette. You know, you be sexist towards your secretary or whatever. That looking back at that type of world today, you're like, wow, can you believe we used to be like that as a society? And I think what we're saying now is just like this idea that you would have this white color work show up to the office like you know type into a keyboard like into your terminal. And like that's the world now. That's going to be the same as us looking at this madman era. I think you're right that it will feel quaint. Well, let's talk about your projects. So you still have this podcast app that you're working on. Yeah. Why haven't you given up on that yet? Yeah. Great question. I am. It's a timely question because I do wonder in this world where we can start building all sorts of custom apps for ourselves. It's very clear that within a year or so, someone else will probably be able to replicate my entire podcast player except make two feature changes because I didn't have time to make their favorite pet feature. And so this world is about to come. So then the question is like, can I even make this a lifestyle business and my suspicion is no. And so this causes me to think about I should probably stop viewing that as something I'm going to make a lifestyle business and view it more as a hobby that I still enjoy quite a lot. And then it's like, well, what should I go do? Right. And I am I think more than ever uncertain because I do not know there are so many things I could do in software purely for the enjoyment of it because I enjoy the production of it. But at the same time, if I think of myself as like a widler that is producing nothing useful for anybody else, but I'm quite quite amused by it. Part of me feels like that doesn't feel like a great way to spend a life either. So part of me kind of wants to produce something that other people value. So I'm not sure. I am right now forcing myself to write at least one article a week on my sub stack and that is much more to force myself to have the discipline to do that. I have never written a single word on that sub stack based on chat Gbt or any LL handcrafted LM. Yes, because I enjoy it. the actual thing, like the actual writing itself. I tried briefly, 'cause I use whisper flow, a voice thing, I tried briefly to dictate a few paragraphs of my sub-stack, and I decided even that removes some aspect of joy. Like for me, the writing itself is a physical activity that I feel like the voice doesn't even feel right when you read what I've dictated versus what I have typed, you know what I mean? So I am like old school, I'm only not old school enough to be handwriting the article, okay, but I'm old school, right? But I'm old school like typing this, right? But I don't know, because I don't want to shift my life into just amusing myself, because I can't figure out a way to produce something that other people will value. But I am readily acknowledging that the skills that I have been valued for in the past might be very quickly going to zero when it comes to incremental marginal value. You should put that on your newsletter. So we'll have this newsletter in the show notes, but like you should just be like 100% human generated. (laughing) I think that label is gonna carry like some amount of like cache A. Totally, plus that sentence needs to have a typo in there. (laughing) You know, I'd be a liar to say that, you know, I don't employ AI from my own weekly newsletter, but I have like severely cut back on stuff, because it's just, you have it right, you know, some amount of the newsletter, then you're like, wow, there is so much AI stink on this text. And so then you're like, I am gonna have this wonderful idea that I will use AI to remove the AI stink from the text, and it can't do it. Part of me is just like, is that a temporary or a fundamental limitation? It feels like it should be a temporary one that I wish it would just go away, but it seems more and more like a fundamental one. I stopped myself and I'm like, wow, I'm asking this AI to be more human and make more mistakes in this stuff, and it doesn't understand. I wish there was a human around here that could help me out with this, and then I like look at myself in the mirror, right? And so I have decided that, you know, with my newsletter, I only use AI for brainstorming and to help me like unlock my creative juices. I wrote an article, you know, "Amazon layoffs were coming, so I decided "I'm gonna write a layoff emergency kit email, "the things that you can do preemptively." Turns out that email came out the exact day that the announced the layoff. So it looks like I was in on it on the inside. No, I did not create the layoff so that my user could do it. - The ambulance chaser. - Yeah, exactly. And so I was like, okay, cool. So I, you know, I started typing out a bunch of notes. I said, like, hey, this is here, the seven things that come to mind. I was like, can you give me some examples of some things that I might have missed? And they came up with two and I'm like, okay, this is great. And then from there, it was human written. And then because I want to move away from AI, trying to remove the AI stink from things, I've literally hired a copy editor. He's a maniac. He might, you know, he's a machine, not an actual machine a human, but I basically say, hey, this article is ready to go and he will knock out the copy editing in two hours, less than that actually and then send it back. So I can insert him into my workflow. - And because he hasn't read Tim Ferriss's four hour work week, he's not outsourcing that job to an AI. - No, dude. And so, you know, let's talk about this. Like, Gurage, he writes the pragmatic engineer newsletter. He went on LinkedIn and he said, hey, I'm hiring for this really interesting job. We at the pragmatic engineer do these wonderful tech deep dives. And so we're looking for somebody that's interested in doing this. And so he listed out all of the things. Like, you have to be a good researcher, you have to understand the tech, enough to ask good questions, you have to have these habits of like, you know, really getting into the details and then you have to be a good writer as well. And the internet unironically was like, have you thought about using an AI agent to do this stuff? And Gurage, like, are you serious? You think I haven't won come up with that idea on its own, but then too, what he's actually found is that the AI has slurped up all of his writing. You know, he has a need for original research. And what the AI is doing is just serving back to him, the stuff that he's created. And he's like, wait a second, this isn't useful for me. To me, this architecture seems like a fundamental limitation of AI. Literally, he's created a newsletter where the value comes from original research, like getting on the phone and we're getting on Zoom calls and talking with people and like trying to find the story there. And so I wrote a little thing. I was like, man, you know, posts, like interesting job because AI can't do the work, internet unironically proposes the novel idea that he uses AI. Well, I think you're absolutely right with your human copy editor and stuff like that, that there's an aspect of which the human touch is still incredibly important. Let me pose you a crazy question about when humans can actually replace something. What do you think the polymarket is on the first year in which an AI wins the World Series of Poker? I think they're going to have to relax the rules for allowing an AI in there. And from what I understand, there is an AI poker tournament. I mean, the interface for a poker player is pretty simple. It's like, bet size, full call race, or flip the table over because you like lost. Anyways, so I mean, at the end of the day, there's a discrete set of actions. Bet sizing is still discrete, right? It's based off of like the chips that you have in front of you. So it's a nice, you know, like bounded problem. But it's also a game of imperfect information. It's not like chess. And so to actually be a good poker player, you have to do an analysis on the people that you're playing with. Are they loose? Are they tight? Are they really good? And so turns out, I was really into poker, you know, and especially in the early 2000s, turns out like, there have been an amazing advances in terms of like poker theory. So there's this stuff called Game Theory Optimization. I don't know if you're familiar with it. So essentially, we have we've exhausted, because it's discrete, we've exhausted all of the games that are out there, right? And so like, hey, you know, what's EV on, you know, this particular situation where you have Ace Queen offsuit or playing a, you know, a game of like, you know, 25 no limit and you're currently heads up. So that is solved in that sense for like what the optimal move is. We haven't solved it, I think, past three players. Because of the combinatorics involved, right? Because of the expenditures. And so I think what's happening is essentially, we have with poker our go moment, where we thought that go was so unbelievably complex that we would not be able to program a computer to do any, you know, then we had like move 37 or whatever it is. I don't know if you're familiar with the AlphaGo stuff. Turns out we just like had it play itself billions of times and then it like learned how to play like a better go game. So my back to your question, the polymarket thing, assuming they let AIs in, which is a big assumption, I think. When would they win the World Series of Poker? I think it's a numbers game at that point. I think that you would essentially, like the most skilled player doesn't win every year. And so basically, you know, if we talk about poker in the early 2000s, you had Chris Moneymaker. He's the first guy to win the big tournament when ESPN was covering it and there was this big moment. Sorry, Chris, if you're watching it, I doubt it. Like you are a terrible poker player. He is not good by and he was bad in 2003. He's worse if you think about like where the meta is moved like today. But he still won because it's a stochastic game. And so assuming they let people in, like it costs $10,000 to get into the tournament. I don't know. I think when more than 50% of the field is AI, yeah, that's when they'll start winning, where you would expect, you know, 50/50 that they would win. And that's the reason why I think they'll never let AI in. Right. Just appear numbers thing. So the reason I like this question and thank you for your answer is because two things. One is I think part of it touches on your comedy question. Right. There's an aspect of in-person human poker play that's different from online poker playing in that online poker playing has far less tells, right? Because you can't tell body language. You can't tell what a person's looking at. You can't tell whether they're sweating. So there's an aspect of it that's like comedy, which is like does the AI need to look at the other players and does it have an advantage if it does so? I don't know the answer, but that's interesting to me, right? But then the other question is, you know how the world's best go player quit immediately after a computer beat a human, right? Well, beat him specifically, but he said he was no longer fun. But yet, chess is being played by more humans than it ever has been. Okay, but chess is a solve problem now. Like you can't even be a chess game on your phone. Okay. I don't know which way poker goes on this. Like, is there an aspect of it where even though humans know that an AI plays better than them, that they will continue to like it because it is much more about winning a poker game is a form of positional good. Like it, like only one of us can win. And it's a strict stack ordering of who's better than whom, right? Like over time, if you play enough games. And so I think that maybe the fun of poker remains because people find that the joy is actually imbiting someone else, not imbiting the computer, if that makes sense. I'll tell you this, the guy that made it that started deep mind. He was a chess prodigy when he was like six years old and or something like that. And I guess he was playing some like Danish like master or something like that. And the game was just dragging out. And then it was supposed to be a stalemated game, but he wouldn't accept a draw. And so like the kid had eventually made a mistake. And so he ended up losing. That was the moment he realized that like, why are we putting all of our effort into these games? Like if you play the best chess game, It's ever existed. The world is still. in the same place that it was before. Right? And so then he's like, "Hey, let's put it into, you know, this wonderful thing called like, you know, trying to get computers to think. I think the jury is still out." But ultimately, like, you can draw a straight line from that, like, "Hey, chess is stupid. Let's put it into AI to beating that go player, right?" To getting to the point where we can fold proteins and like, like, it's going to, like, unlock this, like, new era of healthcare. Yeah. I think all these capabilities and the employment, I think we can all agree whether it's replacement or it's augmentation. There's going to be employment shifts, for sure. Right? I think it'll bring to the fore the question of like, what do we value humans for? Right? And I think to your point about capitalism, for a long time, the capitalist society valued them purely for their economic output. And I think this is going to put that to the test and say, "Hey, do we value things people for a lot more than that?" And if so, the AI is going to make that question the forefront question. So I think there's a related question to your, like, replacement versus augmentation, which is, like, does it drive the economy or does it destroy the economy? And I think we keep flip-flopping between, like, is there collapse in the way that we think about markets? Or is there just a shift with the way that we think about what's valuable and what's not valuable? I think that economic collapse is possible based off of how things are changing, but I think a sane and rational actor can only bet on the market shifting because if it collapses, then nothing matters. Right? And then the world is, like, too different for us to make say anything reasonable. So I think that you have to choose what's in demand, what people will open their wallet for, is going to shift, and that is going to have a ripple effect on, like, employment, on labor, and all of these things. But that's just some moving of, you know, the employment pull around to this new frontier and not a, like, you know, go stock up on guns and gold and, like, water. Yeah, totally. The other polymarket related to that thought that I, myself, I'm curious about, is the first data center bombing. Is that as almost definitely happening, right? The Luddites will definitely do that. So I am curious, you know, where that is. And maybe if I'm an optimist, what I would say is the first data center bombing is probably the moment we can all agree that the writing is on the wall, that the change is permanent, that the doubters are wrong. Right? Did you have to plant that seed on my podcast? Like, I wasn't even heard. You heard it here first. Well, Philip, thank you for stopping in for this conversation. It's really great. I think that, you know, in six months, I'll pull you in again, and we will have a completely different conversation as this landscape shifts. You know what's crazy is you and I last talked, I think, about three months ago. And I would not have even guessed back then the amount of change that would have happened by the time we talked again. So if we really do talk in six months, it will be amazing. I will arrive at my flying car. Well, I, you know, like my probability distribution for flying cars is pretty low. You know, but I do think that like, we will see where this, like, vibe coding swarm coding thing goes. Am I going to be managing nine agents doing this thing? Or am I going to be managing the one agent that's managing the nine agents? Or am I going to be managing the the top level agent that dispatches to the lower level agents that eventually dispatches to the agent that does the who knows where that goes, but it was certainly great to talk through all of this. Yeah. Thanks for having me. Cool. Awesome.