원문 보기
Hey, we're going to have some billionaires tell us about the myths of AI. That's actually surprising cuz I thought it was mandatory for every billionaire on the planet to tell us that AI is everything. So, maybe Yeah, man. We were all over the place. All right, here we go. is the more important point, which is AI-enabled coding, I think, is still 5% of the broad market.
So, it's kind of a nothingburger. Yes, they're leading, but they're leading in something that isn't that big yet. Now, you would say, "How could it not be big?" And what I would say is because most of the stuff that's being written is still white sheet de novo code. And I think the ugly truth is, I don't care what model you have, but the long horizon ability for any of these models to actually build enterprise-grade software is still s h i t And that's the actual lived experience.
Not for me, but when I call on our customers, half a trillion-dollar banks, hundred-billion-dollar insurance companies, none of these guys are like, "Wow, it just works out of the box." It doesn't work. So, most of it is still hand-tuned. So, until I can honestly tell you that we can point a model at this with the right guardrails, which I can't today, what I would say is it's a small market that will become large as these models become better.
But we are in the world where we have 50 years of accumulated tech debt as a world. And I suspect when you enumerate the number of lines that that represents, it's hundreds of trillions of lines of just pretty marginal, mediocre code to bad code. So, I I liked this idea that there are What did he say?
50 to 60 trillion lines of code out there? That's legacy code from 50 years ago, but to for a recap, he seems to be dropping some hard truths on people. Similar to, I think, the hard truths we've been trying to drop on people. And it's unusual cuz I feel like they weren't dropping those hard truths on purpose.
So, let's see. Well, I mean, we still did not give AI complete control over coding. Yeah. [snorts] Well, I mean, you don't. Verdan or Verdan. And I applaud you for that, my friend. Very much applaud you. Oh. But AI is going to hate us by then. Uh Oh, by the time it actually gets big? And in what bubble is he living?
Oh. Wait, so he's saying that it's the AI is s h i t. So, that must be if he's like in the bubble of doom, I guess, rather than the bubble of awesomeness that a lot of the billionaires that I know seem to be living in where AI can do everything. So, but let's let's let him continue and see what else he has to say.
On top of that, we have all these legacy languages. I'll tell you one of our customers, they have to go and get 60-year-old pensioners to come into the office to interpret code No, I'm I'm not joking. This is COBOL, Fortran. This is a hundred-billion-dollar-a-year revenue company. And that's how they solve these problems.
It's not Opus just solves it. So, I would just keep in mind that most of the tech debt in the world that exists, 99% of it is still poorly addressed by these models. We are untying this Gordian knot. It's going to take decades to do it right. So, all the breathlessness about all this other stuff, I really think it's not where the money is.
It's not the big-time stuff. And you can tell me, "Oh, yeah, it's going to be the future." And I would say, "Tell this business that's a hundred billion dollars a year of revenue and 50 million billing relationships that all of a sudden you're going to open-claw your way to a solution." It's Not to say that you can't have a great chief of staff, and not to say you can't do some useful stuff and trickery and, you know, have a good knowledge base.
I'd like that, too. But the core things that your lived experience sits on today is a mess of tech debt that will get very slowly replaced. And that's just the reality of life. Yeah, I mean, I think that that is which makes me wonder, right? Is he trying to intentionally get the AI coding market to bomb so then he can buy that stock and then it'll jump back up, right?
Like [gasps] But he is saying stuff that I say all the time, which is it has huge potential. It's really good for rapid prototyping and brainstorming. But the moment you go into a large legacy project that has hundreds of millions of lines of code, uh if you expect that it's going to be able to integrate that feature and not break a thousand things, try it.
I dare you. Try it. No, I mean, seriously, you shouldn't try it, but if you really if you don't believe me, try it and then hit that push to production button and see how long you have a job. Cuz it's going to be really, really short-lived. I mean, just ask AWS. Cuz I think they literally just went through this.
I mean, hell, just ask Anthropic. I think they literally just went through this. And accidentally open-sourced multi-billion-dollar product. No matter whether they say that was a human or not. I mean, come on. We know the truth. So, I think it does have incredible potential. I think that ultimately it might get to some point where it can do a lot of the hard stuff that we pay software engineers a lot of money to do, which is figure out how to integrate a new feature or address some issue without changing or breaking or destroying the performance or creating security holes or any of that stuff with the existing app.
That's the real challenge. Our boy Chamath Chamath, he seems to be agreeing with us for once, which is strange cuz I thought that just a couple weeks ago he wasn't. Oh, no, his his name is Guys, I got to be honest, I can't say that last name. Chamath Palihapitiya. But he's this pretty wealthy investor that is known for notoriously telling interesting stories.
Some of them that may not be grounded in reality. And there's kind of like an ongoing joke of you should kind of maybe not take his opinion seriously. Ever. But he has been right a couple times, and he was like an early investor, I think, in Facebook. Not as early as I think he says he was, but he was.
So, strange, I I think I actually kind of agree with this. the breathlessness about all this other stuff, I really think it's not where the money is. It's not the big-time stuff. And you can tell me, "Oh, yeah, it's going to be the future." And I would say, "Tell this business that's a hundred billion dollars a year of revenue and 50 million billing relationships that all of a sudden you're going to open-claw your way to a you can't have a great chief of staff, and not to say you can't do some useful stuff and trickery and, you know, have a good knowledge base.
I'd like that, too. But the core things that your lived experience sits on today is a mess of tech debt that will get very slowly replaced. And that's just the reality of life. Well, on top of that, it's more likely than not true that it is a mess of tech debt. And using the AIs that we have uh today, I think that that tech debt is likely to only spiral more out of control, not less out of control, and then make things even more terrible than they are.
Now, I think that you can, again, do collab coding. You just have to be careful and make sure that you're not overdoing it, and you got to make sure you're really careful with your tests and how you experiment with these things and cuz it just Yeah. It can get messy. Sean, I'm going to try to see if I can make this a little bit bigger.
That didn't work. But I think I can read it over here. Ideology saying, "We solved similar problems already, just newer code, like Python, Java, etc. I think ontology again can solve and prove this if done correctly." Well, the part that I think is really tricky is that if you have these banks or these critical systems that are coded in this particular way and they need to be maintained in a way that they don't have any new errors.
That's when I think it gets real sketchy to try to use an AI system because you need to be like perfectly correct on the first try. And with a lot of the legacy languages, they may not be as good. I mean, I know Anthropic has its COBOL thing, so maybe that's something. I haven't tried it cuz I don't actually write much COBOL these days.
As I suspect many of you are not. So, for the legacy languages, I think that probably makes sense even more. And then for the modern languages, I think it's still a problem. And I think it gets harder the more complicated the language. The more it's a like lower-level language cuz there's just like a lot more to reason about.
There's a lot more stuff that can go wrong cuz you have to handle a lot of that stuff. For example, if you're using a language that's garbage collected, that removes a lot of the problems that you have to deal with in systems languages. And instead, you just pay a penalty, you pay a tax for the garbage collection.
But then, the benefit is you get to sidestep a lot of the issues that you would have to solve if you actually wrote the code. So then, if you use an AI in a language that is garbage collected, Java, C#, Go, I think. Then, the AIs I think have uh more of a fighting chance. But when you start to use lower-level stuff, C, C++, maybe Fortran, unsafe mode in C#, yeah, Lisp, I'm not sure.
But I do think just generally speaking, the AIs are not ready to write the entire system, especially for systems that already exist and it's a large amount of code. Cuz I think you quickly get into that devolved state of the ClickHouse uh source code, the ClickHouse repo, where they think it's working, but you can you you and I can see that it's devolving into this really crazy spaghetti code that's not great.
And so even if it works, it's not going to work for long. You know what I mean? Like it works and that's all it does. That that's not the path to victory, guys. That's like proving you can run a marathon by sprinting the first mile. You've still got 24 miles 25 miles to go and if you're already cooked in that first mile, um I don't know if you're going very far.
I mean, the truth is many of the people that I'm going to try to help, they're vibe coders that have come up with these awesome applications vibe coding, which is super which is super cool. I love that. I mean, seriously, I I really love that. We've enabled the global population to start to create software on their own through natural language.
That is a legit right step in the direction that we want to be moving. But then comes the next step, which is iterative refinement, maintaining that code, fixing the bugs, making it faster, making it more secure. And that's when we start getting into these, uh-oh, the AIs can't do this. Now I need a software engineer.
And then it's potentially even worse because, you know, had that software engineer or the software engineering team actually written the code, they would understand all the code. And so fixing that broken code would be less hard. And it's not to say that you can't bring in new people and when new people come in, they know all the code.
They don't. There's going to be some ramping. But historically anyway, by the time you got to a software product that's viable, you probably had one person in the building that knew most of the code. So minimally, you could rely on that expert to be like, "I understand how this works. I can train people and then they can handle some of the stuff." And then that senior person can do the code themselves as well or they can say, "Look, I've done my part.
I'm out. These other people can take over." Whereas now with the AI, if it's written the whole thing, then you bring in that new person. They don't know. And you can ask the AI to explain it and sometimes it doesn't do that bad of a job. And then other times it totally and completely fails. So, so yeah. Good take.
Comments 0