Agentic Coding and the Economics of Open Source

AI is rapidly transforming how software is built, shifting economic incentives from open source code and collaboration toward on-demand, personalized development through agentic coding a.k.a. vibe coding. In this episode, Chris speaks with Miklós Koren of Central European University about how AI is reshaping open source and the software industry. They explore the economics of incentives, evolving collaboration patterns, and what this shift means for software development, the future of AI, and its broader impact on the technology sector. Featuring: Miklós Koren –LinkedIn https://www.linkedin.com/in/miklos-koren-b535a42a/Chris Benson – Website https://chrisbenson.com/, LinkedIn https://www.linkedin.com/in/chrisbenson, Bluesky https://bsky.app/profile/chrisbenson.bsky.social, GitHub https://github.com/chrisbenson, X https://x.com/chrisbensonLinks: Vibe Coding Kills Open Source https://koren.mk/publications/vibe-codingThe Directions of Technical Change https://koren.mk/publications/directionsThe Tailwind story https://www.businessinsider.com/tailwind-engineer-layoffs-ai-github-2026-1Upcoming Events: Register for upcoming webinars here https://practicalai.fm/webinars!

• 0:02Speaker 00

  Welcome to the Practical AI Podcast, where we break down the real-world applications of artificial intelligence and how it's shaping the way we live, work, and create. Our goal is to help make AI technology practical, productive, and accessible to everyone. Whether you're a developer, business leader, or just curious about the tech behind the buzz, you're in the right place. Be sure to connect with us on LinkedIn, X, or Blue Sky to stay up to date with episode drops, behind-the-scenes content, and AI insights. You can learn more at practicalai.fm. Now, onto the show.

• 0:41Speaker 01

  Welcome to another episode of the Practical AI Podcast. I'm Chris Benson, Principal AI Research Engineer. And with me today, I have a guest I've been looking forward to for some time now. I have Dr. Michal Skoren, who is a professor of economics at Central European University in Vienna. And he has written a really interesting paper on the effect of vibe coding on open source. So, welcome to the show. Really excited to have you here today.

• 1:16Speaker 02

  Thank you, and thanks for having me.

• 1:17Speaker 01

  Yeah, so I think this is a slightly different, you know, take for us. We tend to, on the show, you know, leap straight into models and all sorts of stuff. But I know you're a professor of economics, and you study incentive systems. And so, I'm really interested in understanding how you turn that particular lens of economics onto open source. And so, I was wondering if for listeners, if you could, you know, kind of talk a little bit about what drew you into the notion of exploring open source through that lens of yours upfront? You know, what was the first thing

• 2:02Speaker 01

  that said, "This is something that we need to go study"?

• 2:04Speaker 02

  Yeah, let me give a little bit of background on that. So, as an economist, my research is really focusing on competitiveness. So, what does it take for a company to be competitive in the marketplace? Or what does it take for a country to be competitive? And for a long time, I've been really interested in whether it's technology that makes a business succeed, or whether it's a talent that they have, or maybe both, or maybe there's some interaction between technology and talent. And I would also call myself kind of an accidental software developer

• 2:42Speaker 02

  in the sense that economics is a very quantitative science and there's a lot of computational research involved. And but I was never trained as a software developer, but you know, we have to be effective at using your computer. So and and this part I actually enjoy at least as much as as thinking about the economic incentives that you mentioned or the other parts of of the science. And so the the story of this paper is we've actually been thinking with with my co-authors Gab or Aaron and Julian. We've been thinking about the economics of the software industry for some time.

• 3:22Speaker 02

  Um, various aspects and in particular, open source, the open source ecosystem. Why do people write open source code? Where are the open source developers? How do they collaborate in space? I think that's really fascinating that, you know, you can work with someone on the other end of the world. And yet, actually, what we see is that most of the collaborations are highly localized. So they are typically from the same city or or um basically a couple hours' drive from from one another. So these kind of things we've been exploring for for some time when actually AI um came about. And and we initially, we didn't really connect these. So we had the research agenda, thinking about software.

• 4:06Speaker 02

  And then we were looking at AI as consumers. So we were, of course, looking at chat G P T and and the amazing success of chat G P T. And we started actually thinking about, oh, we should be writing more papers. And, you know, I'm happy to tell more about how AI has impacted science. We can get back to that, but let me come back to the paper. So it it it took us quite a while to kind of connect these these two pieces. And so we are actually doing research on this and, you know, software engineering and software development is is one of the use cases where AI is really, really successful. So maybe let's let's try to think about, think about that. For me, the personal, I I very much remember the.

• 4:51Speaker 02

  Kind of what triggered this particular paper, it was like November or December of 2025. And I consume a lot of social media related to technology and software. And so everybody was doing vibe coding and everybody was showcasing the app that they have developed on Reddit or X. And so like, look here, I did this and you know, why don't you download it? And I was just like, my social media feeds were just flooded with these type of vibe coded apps. And I started thinking like, why would I do that? I can do it myself in half an hour. Why would I download your app when I can just go into Claude Code and, you know, do it myself?

• 5:34Speaker 02

  Exactly to the specification that I would need. And so this kind of in my mind connected vibe coding with this idea that you can actually write software for one person or just a few, few people. And then I realized that of course that has a major implication of how we think about the software industry that we have been working on with my colleagues. So I said, okay, let's write the paper about vibe coding.

• 6:03Speaker 02

  And so that's, that's how we started about that.

• 6:06Speaker 01

  Yeah, and it's, and I love the title, which I don't know that we've actually said yet, which is vibe coding kills open source, which is quite provocative when you think about it. And all the components of the title are provocative. I mean, vibe coding, there's been all sorts of commentary for it against and, you know, trying to understand it.

• 6:29Speaker 01

  With open source being very foundational to software coding for the last, you know, three decades at least. And most commercial packages having various open source components to it. With but but that is that is part of a conversation that I know I've certainly had with other people in terms of at this point with these tools, especially, you know, you mentioned Claude Code, and I know that's probably the most popular coding tool recently. I certainly use it. And with Opus 4.5 that came out in November, that really kicked off a lot of innovation. I think a lot of

• 7:10Speaker 01

  us that do code ourselves are wondering why, you know, I can just go do exactly what I want, or I can take existing software and make alterations and stuff with, you know, how how should we be thinking about open source in this new with this new paradigm of vibe coding? Or, you know, it's becoming, you know, kind of a, you know, prompt engineering for software engineering, whatever you want to call it. There's a bunch of names coming out now.

• 7:38Speaker 02

  Yeah, so I think, and we got a little bit of heat because of the title. And I think in

• 7:44Speaker 01

  I'm sure.

• 7:45Speaker 02

  academia is quite conservative. And, you know, this sounds like a clickbaity title, but to this my reply is that it's not clickbait if it's true. But of course, let me explain why we think that is true. And, you know, we thought about every word we discussed in detail, like, is this the correct word to use in the title? And I agree with that sentiment of vibe coding has this negative connotation, but that's, that's the word that people use. So let's just

• 8:13Speaker 01

  Absolutely.

• 8:14Speaker 02

  Let's just stick with that. And so the way we approach the problem, there's, at least at the time, there was very little data. Now we have more data and we have done some new empirical analysis as well, looking at data. But at the time, we were just thinking like, how would an economist approach this problem? And the way I think about economics, and this may not be the canonical undergraduate economics textbook, but that's what I

• 8:43Speaker 02

  Kind of thing that the three main pillars of economics are. One is that people respond to incentives. So whether it's monetary incentives, so they wanna make a good wage. But it might be other incentives as well. So in open source, it's typically not, at least not directly some payment. It could be kudos, if you like, that, you know, it's good to fame. Or it may just be like a good feeling of solving other people's problems and sharing your solution with others. And that's also, you feel good about yourself and feel good about helping out others. That's also an incentive. So economics can also talk about that. It doesn't have to be, doesn't have to be money. That's involved.

• 9:24Speaker 02

  The second pillar, I would say, is that the economy is a closed system. So you, if you take away something here, you have to, just basically have to make sure that things add up. So I think when it comes to AI, one thing that is a very scarce resource, I think, is human attention. So this idea that AI is gonna take over everything and it's just gonna infinitely produce stuff, including software. I mean, it's getting better and better of doing that, you know, longer and longer runs for AI agents. But ultimately, then, has to be a human who instructs the AI and who reviews the result.

• 10:07Speaker 02

  And so there is this attention. There is a very limited resource. So if you turn this towards AI, you have to take it away from something else. And then the third pillar, I would say, is, you know, how these two things come together. So if something is scarce, there's very few of it in the economy, then the price of it has to, has to go up so that people, people can adjust. And so this, this kind of... System thinking in economics actually helps us think about the the open source ecosystem as well, because here, I think, just looking at, you know, individual data sets, or there are a lot of surveys about how people use AI. There are surveys about how software engineers use AI.

• 10:52Speaker 02

  But these are limited, even if you do this on tens of thousands of people or hundreds of thousands of people, they don't necessarily capture the entire system of engineers and users and how they how they interact. And we saw that it's very important to to capture that kind of interaction that as an economist. So actually, all four of us are within economics. We are specialized in initially specialized in international trade, which is fundamentally about this type of market equilibrium forces. But there, there are different countries that are trading with one another. And so here, you know, it's not countries, but there would be a software developer who's who's building a software package.

• 11:33Speaker 02

  And there's a user who is downloading the package from GitHub or whichever package manager. So we and then we started thinking about it and we realized that we actually have the tools in economics to think about this. And we just have to make sure that which of these apply to the open source sector and which of these are are irrelevant. So we do think that the response to incentives is still very relevant, even though it might not be monetary. And so we identified that it's whatever developers care about, it's roughly proportional to the human attention that they get or the visibility that they that they get. So it might be that and and I understand that there are different types of open source projects.

• 12:18Speaker 02

  It might be that the big corporation sponsors a project and it would be different. But say a hobby developer would typically be kind of happier. And actually, even the corporate developers would be happier if there are more people using their product. For the hobby developer, it might be that, you know, it looks great on their CV that they have, you know, fantastic contributions to open source and they might get a better job later. So maybe even earnings wise, they can they can turn it into money and not just. Not just fame, but it might also be the kind of incentives that I mentioned earlier. But in any case, the more users they have, the better.

• 12:57Speaker 02

  So that's one key component of our of our theory of open source is that while it's easy to write, like just a simple project for myself and a dozen people, I'm much more happier if I can share it to thousands and millions of users. And so that's one aspect. And the other is that what is changing with AI is that the technology of writing software is drastically drastically changing. And so that, of course, affects the cost of of writing open source packages, of actually developing the software, but also sharing it, sharing it with others. It becomes much, it becomes much easier. And if you think about it, these two forces actually go against one another in the following sense that.

• 13:43Speaker 02

  So if I have just an idea of of a of a software package, and before AI, I might not think that it's good enough of an idea, or I might be bothered, oh, I have to pick a license and upload it to GitHub and write some documentations because there are going to be users who ask questions. And and so all of these hazards, of course, GitHub itself is

• 14:09Speaker 02

  is responsible for reducing that type of friction for developers. So so now a large chunk of open source packages are on GitHub. And but with AI, I think many of these costs go down. And so it's much easier to produce open source. So this would actually probably create more open source packages than before.

• 14:32Speaker 02

  By contrast, and so this is where I mentioned the this finite resource of attention.

• 14:39Speaker 02

  Every developer, but also users, they are kind of users of open source libraries as well. And so they either pay attention to the to the developer. So just let me give you an example, say, in web web development. And that's actually a kind of a use case that's perfectly solved by by generative AI. That, and you know, we might agree or disagree about different software use cases. But I think, you know, a simple front end development, it's almost hundred percent covered by by recent AI models. And I can ask AI to build me a website with a number of features without ever looking

• 15:22Speaker 02

  at the libraries that are being included there, ever looking at their documentation. And in open source, this kind of feedback is actually very important, not not just looking at the documentation, but if there's a bug, then I can I can report that bug back to the developer.

• 15:41Speaker 02

  And so that type of of visibility of the human behind the open source package is getting is getting reduced.

• 15:49Speaker 01

  Yeah, I I find that fascinating. I that's not something I had not really thought about the notion of the developer's attention being a point of scarcity, you know, in terms of of managing that. But that makes perfect sense when you say that, because I know that I would say everybody that I speak with regularly that is doing development, that is, that has become the challenge of, you know, they can produce a volume of code with the tooling these days, whether it's open source or whether it's proprietary, but they are still required. There's an expectation from their employer, certainly, of going back and validating and making sure that everything is correct and figuring out the bugs and stuff like that.

• 16:34Speaker 01

  And I I think one of the things I'm I'm curious about is how that plays in as we're talking about kind of that notion of what happens to open source from all this. How does that like, what does that imply about the role of the developer going forward? And the responsibility of the developer as you're looking because you know, the volume of code from that, that obviously the, you know, if there's a company behind the open source project, driving it, they're trying to get a lot of users on board and stuff like that. But with these different incentives for different players and that, how, how does that play out? There's a complexity there that I think maybe I missed and possibly other people have missed too, that I think you're delving into.

• 17:19Speaker 02

  I think it's very important. And so you mentioned proprietary software. It's very important to distinguish between the two in that, of course, for proprietary software as well, people like to have many users because they mean paying clients. But the business model, and when I say business model again, it's not necessarily money. But so the model of open source libraries is very different from proprietary. You can be a very successful proprietary software company with, you know, even if you don't have millions of clients, if you have like deep pocket clients, it's totally fine. But for open source, you have to have like millions of users to be successful because kind of the margins are so thin.

• 18:02Speaker 02

  And again, the margins, and there is a great study by scholars at Harvard Business School looking at the value of open source in terms of the value that it creates and the value, the amount of work that goes into it. And the amount of work that goes into it versus the value that creates, there's a, I think at least a thousand fold difference. So, you know, in a business, that would be like a huge, a huge gap. You would mean that you cannot really monetize the value, value that you create. So the only way open source can survive is if you have really, really big user base. And that's, and that's why I think this type of argument that with AI,

• 18:44Speaker 02

  human attention drops, your user base drops, it's particularly affecting, affecting open source. Now, in terms of what are the, what are, do developers do? And, and how are, how are they? So when, when I talk about developer, I think of them as having two jobs in this, in this, in this model, one is to write code and to share their, their code with others. But the other is that there are all kinds of dependencies that you're using. So if you build a website, you're going to install a whole lot of JavaScript libraries. And so you're selecting which of those to to to use based on your familiarity with the package,

• 19:29Speaker 02

  or maybe you go go and read the documentation, which and this is the part that completely. You can still instruct your AI to, you know, go with this versus that because you have stronger feelings about one library versus the other. But you could, in principle, just ask the AI to give me a website and it would. The and so that kind of selection of libraries is actually something that we're looking at in the data now. By now, I think we have sufficient data to try to tease out the the effect of AI in. And we're actually looking at website development because that's something that's kind of easy easy to track. You can actually see what's going on on websites.

• 20:11Speaker 02

  There's this case of Tailwind, Tailwind CSS, which is a very popular CSS library. And and so they had a tremendous increase in in usage.

• 20:29Speaker 02

  A lot of it driven by AI agents, but but also a very big fall in in website visits. And then their particular monetization model really depended on website visits because they had some premium package that they are selling on the website. So if you don't show up on the website, rather new jobs. And so that's.

• 20:48Speaker 02

  So actually, what we're doing in in a new part of the paper that that we're

• 20:56Speaker 02

  working on right now, we have some preliminary preliminary results that I'm happy to happy to share is try to see whether this tailwind story is the exception or whether that's the whether that's the the rule across packages that are relevant for front end web development. And so we did the.

• 21:12Speaker 01

  Yeah, keep going. I'll ask. There's a thing that you have prompted me, but I'd like you to continue and then I will, I'll ask afterwards.

• 21:22Speaker 02

  So here, we actually did a controlled experiment. So this, of course, one of the things that you do in science, but in social sciences, very rare, because how do you control

• 21:32Speaker 02

  the economy? But here, we actually wanted to see what different AI models would do. There is a, you know, there we can control because we can really instruct the AI models. So what we did was the following. We took a hundred websites. It's actually a representative sample of, of very popular websites out there. We described their, you know, what they do, the use case with a product requirements document. And we actually checked, we asked an AI model to check that there are no mention of technology in there. So it's kind of functional requirements or performance requirements, but no mention of any technology, any brand name. So we have basically, but you should think of, say, banking websites or, or car dealerships.

• 22:17Speaker 02

  So the most popular, most heavily used websites, e-commerce has all kinds of different, different websites. And then we asked various AI models across different families and different vintages to try to build that website from scratch. And then we looked at what is the, and really the only instruction we gave is that you have to use NPM to install dependencies.

• 22:44Speaker 02

  Because it's, yes, there are others, but I think it's fairly universal for, for JavaScript dependencies to be installed via NPM. And basically, as soon as you install everything that you need for your website, we actually pull the plug. And then we looked at like, what are the dependencies that the AI model wanted to install? So we already did this for seven, seven models. So now, you know, we could have 700 different websites, but we actually pulled the plugs. I think we do wanna, do wanna build some of these just to see kind of how, how different they look from the actual websites that, that were the seed of the, of the experiment. And then for every model, we know when they came out, like when they were released. So we could see.

• 23:25Speaker 02

  And there's a lot of correlation, of course. So Tailwind is very popular across all these models. So it's almost universally recommended to be, to be included, but not. So some websites need a calendar, some websites need a chat. Box or or a map and so these features are are different and the different models have different opinions about this and so we could track kind of how uh when when a certain model started recommending a package. And then we look at uh two uh outcomes that we can actually measure in the data at the weekly frequency, we could go deeper, but I think weekly is is sufficient here. One is downloads from NPM where where these packages uh more frequently downloaded.

• 24:09Speaker 02

  And so that, of course, would include the uh demand generated from uh generated by AI, AI models, just go and download, download these um packages. And then the other metric is uh stars on GitHub, which we think of as a proxy for human attention. Uh so you really like a uh a package and say, okay, I I really wanna engage or at least uh, you know, just show that I like this package and I make the effort to go on GitHub and and and give it a star. And so uh so what we find, and actually it's very much in line with the predictions of the model, uh the first one may not be very surprising is that as soon as uh more and more models start recommending uh a library for uh, you know, some of the one hundred

• 24:52Speaker 02

  use cases that we that we have, uh downloads go up. So for every additional use case for which there is a recommendation, downloads go up by like three to five million per week. Uh so it's a pretty uh, in terms of percentages, it's something like uh three, four, five percent uh of of weekly downloads for the typical typical package. Um but by contrast, what you see in stars is that um uh they often actually go down. So um at the very least, they don't go up as much, uh they are the zero or or they actually fall. So so for packages that kind of become uh very, very wide coding friendly, uh at least

• 25:34Speaker 02

  the mechanism and the model is that you divert attention from humans towards the machines. And so the machines are downloading, but the machines are not interacting with the developer on GitHub. And so that

• 25:46Speaker 02

  That already seems to be. So this is data for

• 25:51Speaker 02

  for basically 2025, where I think a lot of the agent software revolution was happening. And so already in the first year of that, you can you can see this effect.

• 26:01Speaker 01

  I'm curious, if you extrapolate out, you know, the the use of agents selecting, you know, these different libraries for inclusion, especially since you very specifically did not constrain that in the prompt upfront, you gave it the choice of doing that. If you were to kind of take that out to the extreme case, is there

• 26:27Speaker 01

  given the fact that a lot of the, not only the libraries, but a lot of the tooling, you know, the the very notion of version control and such are really human constructs that, you know, we we humans historically are trying to deal with complexity as we're writing code. And we have created all of these ecosystems of toolings and libraries and stuff. You know, you mentioned Tailwind, you know, Tailwind exists because it makes it easier to implement CSS for a developer. And it's a very friendly, I've used it myself a lot. And and so, but you can, you know, you can look at all of these different tools and libraries out there in the same way.

• 27:07Speaker 01

  Is there a case when you're when you're assuming that you have kind of unlimited prompting available to you on a on a very high capability model with agents such as, you know, Opus or something, is there in the future, is there a need to have libraries in that, you know, when you talk about open source and, you know, going back to the title, is there a reason that the model needs to use a Tailwind and other, you know, other capabilities to integrate in with it rather than just construct it based on those functional requirements upfront? Yeah, like what, how does, how does the the notion of using these existing things that humans created already

• 27:49Speaker 01

  versus the AI just going and saying, the functional requires this. And, you know, here's all the requirements I've been giving, I'm just going to produce what what it is from scratch without the notion almost as disposable code, because if you needed to make a change, you might potentially just do it again, assuming that you can get some consistency.

• 28:09Speaker 02

  Yeah, that's a great question. And so I can reflect a little bit on my personal experience by coding while I'm doing like scientific computing. It's a very different like I I I spend a lot of time.

• 28:25Speaker 02

  You know, instructing my computer, but it's of course very, very different from what it looked like, say, two years ago. And and it's not so much so for me personally, the the big difference in in how I how I program is not is not so much that, OK, now English is the programming language. I think it's actually it's strange at first, but it took me like two weeks to get through it.

• 28:52Speaker 02

  Then it took me some more time to realize that that structures are still quite important. So you cannot and it might change with different different generations of AI models. I remember it was, say, about two years ago. So it was it was more the the tab completion era and not the fully genetic era. But it was an aha moment where I was I was working on a scientific paper and I and I and I started writing some code in Julia, which is a scientific programming language. I I love the language because it has a very good high level in interface. So you can it's very easy to get started as a scientist who knows little about programming. But it's also a proper programming language with all the right

• 29:37Speaker 02

  abstractions and and tooling. And I set up. I I wanted to solve a problem and I set up the the function properly. And and as soon as I typed out the full signature of the function, I gave it the proper name. So a lot of scientists, when they engage in scientific computing, they would they would be very sloppy about these things. And I'm going to call my variable X, my other variable Y, and and these kind of things. But this, you know, the function name was a verb. There were meaningful arguments. They had. Different types and they were, the types were declared, which is optional. But I thought I was gonna declare, you know, what I'm expecting. And as soon as I finished with the first line, Copilot threw the entire function in like

• 30:22Speaker 02

  one shot and it was perfectly fine. Because it understands what I wanna do. I didn't even give any instruction, I just had to push that. It understood what I want to do by kind of looking at the names and the types and figuring what the types represent and then. So this is gonna be a binary search problem and so I'm going to have to do this and immediately figure this out. And so

• 30:46Speaker 02

  the way I think about AI in, in, in software, but I think in a lot of other use cases in, definitely in science or other knowledge work, is more like a very, very capable, very fast coworker.

• 31:03Speaker 02

  As opposed to, as opposed to some machine or some tool. And I think that, that's still a little bit hard to get, get used to.

• 31:16Speaker 02

  So in science, there's a lot of discussion about how to use AI, how to be efficient and increase your productivity by using AI. And I, I feel that some of the discussion is misguided, as if it were like, oh, you know, can you share with me your skills for, you know, writing a research paper, you know, running a statistical analysis? And, and so you basically, you, it's like your colleague. Like, should I share my skills for talking to colleagues in the coffee break? And like, no, you just walk up to them and you tell them what you're working on and hopefully they're gonna have some useful feedback. And so it takes some getting, getting used to, but I think that, that type of, of, and

• 32:01Speaker 02

  of course, much like with, if you're working on a, on a project with a colleague, you, if you have good structure and good coding practice, it helps you collaborate. And so if anything, I think the importance of, of, you know, maintaining good structure is more important than before because you have a very effective colleague who can immediately.

• 32:23Speaker 02

  Do what you ask and give you feedback. And, you know, the better your practice is, the more, more you can get out of them.

• 32:30Speaker 01

  So that i- it enhances that collaboration and communication between you and your, your AI colleague in this case. In terms of being able to, to have those structures that exist for the, versus everything from scratch. Is, am I understanding you correctly on that?

• 32:46Speaker 02

  Yeah, I'm sorry, I would, I would think that this is primarily a kind of a collaboration problem. Now, the, the question you, you asked earlier, so maybe there is the next level where, you know, a lot of the work are done by AI and maybe they don't need this type of, of interfaces. And, and maybe they would, you know, maybe just write, write things from scratch. Oh, there's, yeah, sorry, I started saying like, what, what is the, so, you know, programming in English is kind of weird at first, but you can get used to it. But what took me more time is, is the type of programming that you're, you are mentioning, like throw, throw away code. Typically think of, I have to design the thing and then, am I going to use it two times? Then I'm not going to abstract.

• 33:31Speaker 02

  I'm not going to write a function or, or, or do a bigger abstraction, I'm just going to write it.

• 33:38Speaker 02

  Am I going to use it 10 times? But with, with Gen AI, you don't need to do, you, you solve a problem. And in like scientific computing, I think that's actually a good practice to just, you know, go. It's very iterative process, you, you never know what the data is going to tell you. You know, let's just try to do this. And then if you're kind of happy with the result, okay, can you say this is a skill? And so now you have a computer program. And then if you do something very similar the second or the third time, one of my secret problem that I, I use quite often, several times a day is, is just the following. Do you want to update your skill? It's okay to say no. And so quite often the agent would say, I'm fine with my current skills, but sometimes, okay.

• 34:17Speaker 02

  So in the last chat, I learned that you like it this way or that way and would update how to approach that particular problem. So, so this idea that you, you, you started a very concrete implementation and then you kind of generalize from there is, is I think very. Very new and and I think it could have this effect what you're what you're saying that besides a few core libraries that you don't really need need anything else. And it's kind of similar to how the these agentic harnesses have evolved, like, you know, what can you do with the model? I remember when there was the MCP fashion, oh, I really want to get into MCPs because

• 35:01Speaker 02

  they sound like really cool and and it would be so useful for me because I use all kinds of different tools and if only I could I could orchestrate them all at once. And then this was actually before I started using Claude Code and then I realized that I'm actually using the command line a lot of times anyway, and you can do everything on the command line. So it might be a little bit inefficient. So now I basically don't use any MCPs at all. If you have a command line client, just figure out, you can ask for help, figure out how to do it and and and they would do it. And so that could could be kind of a next next level of of agentic programming that these agents have.

• 35:43Speaker 02

  I mean, it would still need a handful of tools, but it doesn't doesn't have to be yeah, doesn't have to be a lot. But then they would have to be really, really properly maintained to make make sure that. So that's that's one issue with open source is that, you know, if we froze everything like today and and nobody ever contributed to open source anymore, it wouldn't mean that we are staying stuck. It would mean that it declined. So you need to maintain it because there are unknown bugs, there are vulnerabilities. And so it's a very complex. I mean, ultimately, every software is very, very complex. And if you don't know about these vulnerabilities and there's no one to to fix them, then then the quality of open source would would decline.

• 36:28Speaker 02

  So that was kind of the motivation for us to be, you know, kind of a little bit alarmist when when giving a title to our paper.

• 36:35Speaker 01

  Yeah, well, you know, and you you produce this paper at a point when Really, you know, every organization on the planet and the people that work, both the both the technical staff and the, you know, management of of the each organization is trying to to wrap their head around this, you know, really rapid change that's happening and and figure out the implications. Do you have, do you have any thoughts on how people should be thinking about this? You know, as they're looking at, you know, they're trying to figure out what they need in terms of programming staff, and we've seen some companies announcing layoffs and, you know, going through that, that's in the news.

• 37:17Speaker 01

  And and others are are only maintaining their senior staff, you know, and a lot of the college grads these days are struggling to get into jobs. And people are trying to understand how does this fit? You know, if you look at the last 20 years, and I'll just kind of make up a an arbitrary example, it's very common for organizations to be participating in and building on open source as a foundational material. But given this relationship dynamic that you've just described in terms of how things are changing, if people that are watching this right now are listening to us, you know, how might they be thinking about this based on on what you're looking at, you know, right now?

• 38:00Speaker 01

  Like, what what changes would you say, you know, what where are they falling behind? And what would you say? And I'll let you take that any way you want, whether it's from a managerial perspective, a developer perspective, a product perspective. I'm just curious what your what your thinking is on this.

• 38:16Speaker 02

  That's a very broad question about kind of AI and the labor markets. Let me let me begin with the with the software part, because there, I think, at least we understand what AI is currently capable of, and it's very capable. So so I think if we understand how this works in software, I think it helps us understand maybe the other other parts of the economy as as well. And so the way I think about a software engineer's job is that there's at least three different things that you should do as a software engineer, maybe with different degrees, and maybe you can add add some more.

• 38:56Speaker 02

  But so you need to understand the user, and maybe not, maybe it's not in the same person. So maybe. These are different people in the company. But so you need to understand the user needs and kind of translate them into, you know, not necessarily actually decided not into a program yet, but just to kind of figure out what they really need and kind of what's feasible and what's not feasible. And then then kind of designing a system, you know, the different components and how they how they come together. What are the relationships? And then finally, you know, writing the actual code in whichever, you know, syntactically correct Python code.

• 39:37Speaker 02

  And and I think this last part is basically out, like it's hundred percent automated. It helps if you can still do it and review review code. But I think the fact that it's out doesn't mean that, you know, software engineering is over because I think the the first two are actually not very easy to do. If anything, they are harder to do. And of course, with more and more capable models, the design, you could sometimes let the AI do it and they can. I mean, one thing is that they will almost never say no. So they always pretend to to complete a task and but it might not be a design to your liking or design to your specifications.

• 40:20Speaker 02

  So I think the thinking part you cannot really you cannot really get rid of. And I think in particular, the the interaction with the users and figure out what they figuring out what they really need. So the AI can build you a website, not like what should it be? It can be build any website. So someone should really figure out what are the things that are important and that are not important. Now, the challenge, I think, which is, of course, it is still a. So so I'm kind of optimistic in the sense that I think of AI as a as a tool that augments productivity of of knowledge workers and not really replacing them. Because I do think of knowledge workers as more than just translating into, you know, syntactically correct Python strings. I think their primary job is to think.

• 41:03Speaker 02

  But there's a lot of bundling in school and also in jobs that, you know, we think you're a good developer if you write good Python code. Or or, you know, we teach programming in various programming languages. I think it would be really, really interesting to think about, how do you teach programming when the programming language is English? Um. Because I still think you need to, you need to understand this computational thinking. Um. To, to be able to, you know, interact with, with your AI, with your AI agents. So I'm, um. You know, it's not a very concrete forecast or recommendation, but at, at least in spirit, I'm, I'm optimistic that, uh,

• 41:49Speaker 02

  that, you know, I think, I think we have, and this is my trade, trade economy, international trade economist, uh, speaking, one of the key results in, in international trade is the result of comparative advantage. And that goes back to David Ricardo, that even if you take two countries and even if one country is uh absolutely more productive in everything that they produce, you can still gain uh from trading with them. You just do what, um, what you have relatively more, more advantage in, even if you're kind of absolutely disadvantaged. So even if AI can do everything better, um, we can kind of exploit. And I think the thinking part will always be, will, will always be, you know, a human comparative advantage, even if AI

• 42:32Speaker 02

  can do a lot of thinking itself. And the reason why this comparative advantage story is important for AI is exactly the, the resource constraint that I mentioned at the beginning, that, uh, you know, I could ask Claude to, you know, build me a website and it would go on for a day and build me whatever, but uh I would have to spend my own time to, to look at it, think about it. And so uh it, it, it's actually not a good use of my time if I don't spend it thinking and, you know, I, I, I would trust Claude to do the thinking. So in my scientific computing, I, I actually, my, my, um, my work has changed a lot. I do more, more computing than before, almost no uh writing of code, like practically zero.

• 43:16Speaker 02

  I do a lot of thinking and actually decidedly on very analog tools like uh pen and paper. Or, you know, I went back to reading books about uh different methods to kind of step away from the computer and, and, and think more about the, more about the problem. And then just kind of because also the translation barrier has has practically disappeared. So if I if I can kind of semi-coherently talk, talk about the my ideas that I came up on a walk in like a fifteen minute voice recording, that's good enough to turn it into working code. So and I think we should, what we should be doing. And and this is what I tell tell my fellow scientists as well.

• 43:59Speaker 02

  When you're asking like, how should I, how should I work with AI? Well, you should figure out, kind of try it and see how how it works. And then try thinking about what is it that you want to focus on? And you can basically outsource everything as to, at least a lot of the tasks you can outsource to AI. But it doesn't mean you you should outsource everything. So kind of keep, keep the core scientific activities to yourself.

• 44:25Speaker 01

  I think that's great advice. I guess as we wind up here, I I am curious. I'd like to ask you to step out of of the firm research that you've done so far. And kind of think a little bit about where, you know, unscientifically, you know, just speculative. You know, when you're, when you're kind of done and you're thinking about the future. What are your thoughts on where things may go on this and what might evolve? And recognizing that this is purely prediction and may not play out that way. But I'd love to hear kind of what your personal view is. You know, about where things may go in in terms of things you've not yet had a chance to research.

• 45:06Speaker 01

  And maybe things that that evolved that you don't even know that you would be researching yet. Just kind of playing out the timeline a little bit at to some arbitrary length. Can you share some of those thoughts?

• 45:18Speaker 02

  Yeah, I I think one, and so my thinking is still very much influenced by be being an economist. But these are not research projects. These are like vague ideas of of thinking about the future. These are really, really exciting times to to be an economist as well. Because these are kind of big systemic changes. Like, you know, we've had this type of technological changes, but not as rapidly. And and it's really, really fascinating to to live in and and think about. And one, one thing that I haven't seen, I haven't really seen mentioned a lot of time. But I think it's a very important feature of of at least the current flavor of AI is that it can be very localized.

• 46:02Speaker 02

  So so by now you could actually buy, you know, one of these fancy boxes that have a very high performance GPU in them and it would be not much bigger than a laptop. Maybe it would be a little bit thicker and of course it would cost much more. But like no internet, nothing, it would and it's actually already good enough to run fairly good open source models, which are not state of the art, but they are basically of the same quality as as like the state of the art was last summer. And so I really like to think about, you know, how much different that economy could be when everybody has kind of knowledge locally available to them.

• 46:45Speaker 02

  Because a lot of the digital economy is really built on platforms, so like Google and Facebook and similar companies and and they made a killing of just making sure that everybody's connected to them and they are kind of a gatekeeper and and if you have to go through them, they set whatever price they they want. But I think this force of of intelligence be becoming very, very cheap and even locally reproducible is a is a force that goes the other goes the other way and that could be

• 47:25Speaker 02

  just completely rewriting what we what we now understand about software, digital economy, knowledge industries, basically the entire economy and society.

• 47:38Speaker 01

  That that's a big way to end right there. That's a so a really, really interesting conversation. McLush, thank you so much for joining us today. I hope that as you continue to research these areas, you'll come back on the show and share some of the ongoing research that you have in the future. But definitely given me quite a lot to think about and some new angles on that that had not occurred to me before. So really appreciating that that economic lens on on this on, you know, this topic. So thank you so much for joining.

• 48:16Speaker 02

  Thank you for the opportunity.

• 48:23Speaker 00

  All right, that's our show for this week. If you haven't checked out our website, head to practicalai.fm and be sure to connect with us on LinkedIn, X, or Blue Sky. You'll see us posting insights related to the latest AI developments, and we would love for you to join the conversation. Thanks to our partner Prediction Guard for providing operational support for the show. Check them out at predictionguard.com. Also, thanks to Breakmaster Cylinder for the beats, and to you for listening. That's all for now, but you'll hear from us again next week.