Transcription
What would you advise those who are just starting to learn to code? I have Thomas from GitHub, the largest platform for developers in the world with an audience of over 150 million users, visiting me today. It was under his leadership that GitHub Copilot became the most popular AI tool for programming worldwide. Currently, large companies are pausing hiring. In a couple of years, I won't need a developer at all. >> Thinking that AI without coding knowledge will allow you to build a billion-dollar business is a mistake. If it were that simple, everyone would be doing it already. >> He conducted the sale of GitHub to Microsoft for $7.5 billion, and now he is changing the future of programming. So, aren't you afraid? >> This is the dream of any developer. To wake up on a Sunday morning with an idea, and by evening, to already have a ready-made application on your phone. >> Hello everyone. Today we're talking about vibe coding. We are at the Vivatech conference in Paris. And today I have Thomas, the CEO of GitHub, visiting me. I'm very happy to talk with you about what's happening in the world of development right now. But first, let's figure out what vibe coding is. >> Oh, that's a difficult question to start with, if I can put it as simply as possible. Vibe coding is when you open your IDE, for example, Copilot or Cursor, or any of the similar ones. And then you switch to agent mode and give it a task. And then you just follow what the agent suggests, right? That is, you run commands, you communicate. The focus is no longer so much on the code itself, but on the dialogue with AI. And you don't even always understand, right, what exactly it wrote. You just check if it works. >> Yes. And you don't need to learn to code to try to create something. I tested GitHub Copilot, just by chatting with it and asking: "Make a website, make this, make that." And it wrote code for me and told me where to insert it so that everything would work. How complex can a website developed with this approach be? Can I build something with a database, or is the maximum a landing page or a simple application? My rule is this: you can go as far as your patience for giving prompts allows. That is, if you don't understand what the agent is writing, what the code looks like, then the only way to control the result is to figure out how to formulate prompts correctly. It turns out to be almost like a quiz or even a game. You try one approach, then you try another. I like to compare it to image generation. When you first make a simple prompt, for example, "generate an image of Paris," and almost certainly get something that doesn't meet your expectations, then you start refining your request. And for a while, there were life hacks in Stable Diffusion and Midjourney, for example, adding the phrase "Trending on ArtStation." After that, the result improved. But at some point, you either reach a dead end where you can't move forward, or you simply accept what you got, transfer it to Photoshop or code, and now you have to start learning to program. So, I think you can create a web page, authorization, settings, and so on, but sooner or later you will reach a level of complexity where you will need to understand what the code actually does. Or you will create something that doesn't scale, and everything will work very slowly. And you will have to figure out how to speed it up yourself to explain it to the agent. For example, you have a Shopify store, and you need it to withstand the load on Black Friday. That's where you need to be a professional developer, at least for now. >> Currently, with vibe coding, we are doing some basic things. And how do you think, in 2 years, will it be possible that I won't need a developer at all to build a billion-dollar company? Or is it too early to think about it? >> I think that if you work in a tech business, you still need to be a developer, because whatever you can do with AI without programming, everyone else will be able to do as well. And, consequently, your business will no longer be different from others. Right? If I can just write a prompt in 5 minutes and do everything myself, then why do I need anyone else? Why do I need SaaS, for which I pay a subscription? Therefore, I think that startups will create things 10, 100 times more complex than they do today, with the help of AI. And this is what will become their advantage. Now, you know, there are many businesses where you don't need programming at all, for example, your YouTube channel, and there are many other channels where you can build a brand and then hire a team to do most of the work. But I think the idea that AI without any programming skills allows you to simply build a billion-dollar business is a mistake. Because if it were so, everyone would be doing it. >> Well, yes, and then everyone would have billion-dollar companies, >> which doesn't work. >> But it doesn't work like that. To build a billion-dollar company, you need not only ideas and technology, but also a strong management foundation. Because even the strongest products can be on the verge of failure if everything in the business is chaotic and unmanageable. Operations spare no one. Take Nvidia, for example. Today, it is the global leader in artificial intelligence and the most valuable company in the world. In July 2025, they were the first in history to surpass the $4 trillion mark, overtaking Apple and Microsoft. And in the nineties, Nvidia was a small startup drowning in debt, with failed releases and fierce competition. It was at this moment that they made one of the riskiest decisions in business history. They terminated their largest contract with Sega to focus on their own breakthrough chip. What I'm trying to learn is to discard some projects to focus, but so far it has given them time and a chance to perfect the series, and then the legendary GeForce. Later, they bet on CUDA and a market that simply didn't exist then. AI computing. All these decisions were made not at the moment of success, but at the moment of despair. But today, NVIDIA is the most valuable company in the world. What helped them? Not just technology, but strong operational management. Those who know how to allocate resources correctly, focus, cut out the excess, and build a system that works, win. If you are a manager, a project manager, or were the right-hand person to a company founder, you already have a foundation to grow into an operations director. And this, by the way, is one of the most in-demand professions in the market today. On Headhunter, there are over one and a half thousand vacancies, and salaries start from 250,000 rubles. If you want to reach a new level, pay attention to the Operations Director course from Eduson Academy. The program includes 320 interactive lessons, 49 business cases and practical assignments, and a module on neural networks for business. You get lifetime access to the materials and 365 days of curator support. With the promo code Mogil, you get a 65% discount and three free courses of your choice: SQL, Analytics, Soft Skills, Excel, and others. And if you apply through the link, you will immediately receive three templates of key financial reports on profit, cash flow, and management budget. By the way, the course is licensed, and you can get a 13% refund through a tax deduction. If you are ready to make a turn like Nvidia's, start with operational management. Link in the description. Activate the promo code Mogilko and take action. What do you think, in 5 years, will there be more or fewer developers? >> I think there will be many more developers, because it is much easier to learn now. We already talked about children. They can just open an app and write how to make a game. >> When they, for example, go to a ski resort and there is a Nintendo corner. And therefore, naturally, when children start exploring these technologies, they want to learn them themselves. And if you give them an assistant and an agent who will prompt: "Hey, look, this is how you can learn programming. This is roughly how you can fix bugs. After all, the most frustrating thing in learning is getting stuck at some point, and at home or among family and friends, you have no one who can help you with this, because everyone is not a tech person. This is democratization through AI. Everyone who wants to can learn. This doesn't mean that everyone who wants to learn programming will necessarily become a professional developer. But, as it seems to me, there will be many more user developers, those who do things for themselves, create their own micro-applications, personalized things, for example, applications for a trip to Paris to determine which places to visit or to collect photos that we have taken. This is valuable for you, your family, friends, but there will also be pros who create complex systems, like these agents that we see at the exhibition now. I think the profession will still remain. And there will be companies that will be smarter than others and hire more developers. Because if you increase the productivity of one developer tenfold, then 10 developers will be able to do 100 times more. >> It feels like the way programming is developing now resembles the leap that happened when everyone was making their own websites 10 years ago. And when Squarespace appeared, everyone had websites, and designers were needed, not necessarily with technical skills. It's the same now. Everyone can make an application for themselves, but someone will still be needed to monitor that application, update it, add functionality. >> Yes, and you wouldn't start a business creating websites for small businesses now. Any venture capitalist will tell you that it's not scalable and it's not unique. Therefore, I think that artificial intelligence will generate much more ambitious ideas, but they will be realized by smaller teams. Or you increase the team and scale faster than those who are simply saving costs with AI. >> It's very interesting that we're talking about this now. Because when I share this thought with my subscribers, they say, "Teams can grow, they can do more," and they say: "But who will buy all this then? There are no more people." How do you look at this? >> I think it's a temporary effect. Naturally, companies are now trying to maintain stability and just understand how the market will develop further. But I think very soon people will appear who will say, "Wait, if I have one productive developer, why shouldn't I hire another one and another one?" And actually, you know, AI has already increased the backlog of tasks. I haven't seen a single company that said they've cleared their entire backlog. They had everything done. And as soon as AI becomes even more powerful, all ideas will be implemented immediately, and we will just sit idle. I think, in reality, with all these AI models, agents, the path to AGI ultimately means that we will have even more work. I was just talking this morning that 90% of all code will be written by agents. And it sounds like we are taking 90% of the work away from developers, and they are left with only 10%. But if the total volume of code increases tenfold, and now an agent writes nine parts, then the developer still has the same one part that they had before. And if we continue to reason in this way, it becomes clear that ultimately, the companies that succeed are those that use AI to accelerate processes, not to reduce costs. >> But at the same time, we see large companies pausing hiring and telling their employees: "First, figure out how to solve the problem with artificial intelligence, and then hire a new person." >> I think this is a reflection of a rapidly changing market and great uncertainty, both in the political and technological spheres. It's logical that the natural reaction is to slow down a bit, and you might also decide that some employees are no longer suitable for your company in this environment where everything is changing at an incredible speed, where you are almost forced to use AI to keep up with competitors. As it seems to me, in recent months, many companies have realized that if we have employees who say they don't want to use AI, it ultimately means that as a company, we are not ready for the future, because all our competitors are already using AI and making it mandatory. And in my opinion, we are currently going through this transition phase. But I believe that very soon we will see acceleration. And we are already seeing, for example, Mark Zuckerberg attracting the Scale AI company. I think this shows how Mark thinks and where he believes the future is heading. He is willing to invest in it. What would you say to those who are just learning to program? What should they focus on to definitely find a job? >> Learn together with artificial intelligence. I think the main advantage of young people is that they master new technologies much faster than those who are working day in and day out. When you have a job, you have so many things to do and meetings, so many emails and everything that needs to be done, that there is almost no time left for learning. Whereas young people who are still studying in school or college have a lot of time for learning. And often, you know, young people are much more open to exploring and mastering new things. It's like when you approach your parents and say, "Well, you're old, because you listen to all this music from the past and watch regular TV and things like that." I think the next generation of developers will grow up with AI, just as Generation Z grew up with smartphones. And in my generation, I didn't have a smartphone until I was 20. More precisely, I didn't even have a mobile phone until I was 20, and a smartphone only appeared when I was already over 30. I think we will soon see a new generation of software developers for whom using AI agents in their work will be completely natural. They will use them when writing an email, when planning a trip, and even during the trip itself. And, of course, in their work, not only in programming, but also in all other aspects of professional life. Do you think everyone should try vibe coding now, or is it too early? >> I think now is the perfect time. There are already enough tools, like ChatGPT, Claude, and many of them have incorporated elements of vibe coding. >> Well, yes, you just need to copy the code and paste it, right? >> Not necessarily. OpenAI has Codex, which allows you to do this directly within the Open AI environment or ChatGPT. But you're right, there are tools where you still need to understand how to deploy a project, how to host it, and generally know what a GitHub repository is or what GitHub is in principle. There are also technologies like Vercel, Replit, Glitch, Bolt, where you can really start without any technical experience. If you follow the threads on Reddit, you'll see that you'll definitely get stuck somewhere. That's the point. Either you don't ask deep enough questions, or you don't ask enough problems. Then, perhaps, something works out for you, but it looks bad or too simple. Or you will really reach the limit of possibilities, and then you will get stuck if you don't know how to go into the source code and make changes. Yes, I think Manos is another example. Manos is a Chinese startup that brought to market an agent with which you can do coding from the perspective of an ordinary user. The user specifies what they want to create, and then this agent creates a web page or web application for you directly in the chat. That is, you don't even run anything, you don't deploy, you don't do any development actions. You just ask it to make, for example, a tracker for your child's pocket money or for a trip to Tokyo. Or, you know, such personalized applications that are useful only to you. For example, a workout tracker. Before, you would download such an application from the App Store. >> Interesting, and do you believe that soon there will be a time when artificial intelligence will generate ideas better than us? I think AI can help us generate better ideas. It's a powerful tool for self-reflection. For example, you upload your notes to an AI system like ChatGPT and ask: "Hey, what am I missing?" Or: "What else could I think about? Or if I take this and combine it with that, what will be the result?" That is, I think that the ability to maintain a chain of reasoning that these AI systems possess, combined with your own ideas, your triggers, those things that keep you up at night and that you are so passionate about. That is, all of this still comes from a person, but AI will help us explore these ideas, format them into a pitch deck. >> This is if we look at the next 5 years. And if we look towards AGI, do you think it will happen by 2030? >> I don't know. I think there are arguments on both sides. On the one hand, one can say that the models that underlie these AIs today are smarter than what you and I are capable of. At least, if we talk about the amount of knowledge, right, >> they store a lot of information. They can reason based on this knowledge, they can even summarize information in a very short time. You wouldn't be able to read a 500-page book and retell it. So are they smarter, or at least more capable? Artificial intelligence is already familiar with all the main ideas, >> but they are not creative. I think emotions play a big role in creativity, and they don't have emotions, and they don't have self-awareness. And if AI or AC is defined precisely like this, I don't know how long it will take, because we haven't seen any research on how to integrate emotions into an artificial intelligence system. So we have a way to go. You know, Waymo in San Francisco is amazing. And now these technologies are coming to other cities. Self-driving cars feel like AGI. I think vibe coding also feels like AGI to some extent. But if we consider human intuition, the ability to collaborate, and that "aha!" moment, "let's do something completely new," then I don't know how far we are from that, but it's definitely not tomorrow. >> So, you're not afraid, or is there still fear? At what level is your anxiety about all this? >> I'm not afraid at all. I still have to remind my children three times to unload the dishwasher. There isn't even a prototype robot that can do that yet. It's too early to worry. It also seems to me that all this will happen. Robots are already walking around Silicon Valley, >> but they still can't get a plate out of the dishwasher. >> I notice how quickly everything is developing, and I think, "What should I teach children now? What do you tell your children? What should they prepare for now to be in demand?" >> I grew up in East Germany before the fall of the wall. I was 12 years old when Germany reunited. And I tell my children: "Listen, you are living in such a cool and exciting time. And there is so much technology around you. Today you can start a company not only in a garage in Silicon Valley. You can start a company in a garage anywhere in the world, because all you need is good internet, a laptop, or even a mobile phone >> and a developer." >> And a bit of vibe coding, but you can become a developer yourself if you want. Now anyone can become a developer. They no longer need to seek access to books, magazines, and computer clubs that are only available in their city. And therefore, from a developer's perspective, now is the most exciting time for programming. This is the dream of any developer. To wake up on a Sunday morning with an idea, and by evening, to already have a ready-made application on your phone. And the reality today is: I have an idea, and I start racking my brain. How to turn this task into source code and libraries? How, for example, to make rounded corners in an iPhone app, and things like that. And by evening, in the end, nothing is really ready. Essentially, I just sketched out a project, set up a few things, and that's it. >> So, you're essentially teaching them entrepreneurship? >> I teach them to explore the world. I teach them to solve problems independently. I teach them, or if they learn themselves, how to use AI. And I think it's important to be curious and open. You are probably one of the most positive people I have spoken with about artificial intelligence. Many are afraid, they say, 90% of professions will disappear. >> I'm German. >> And you're German? That gives some confidence. What would you advise those who are still afraid that artificial intelligence will take their jobs? I think the best way to deal with the fear of losing your job is to embrace new technologies, learn them, figure out how to use them, and improve yourself for the profession that will be in demand in the future. If you are afraid that your job will be replaced by AI, then the best way is to become an expert in working with this system yourself, because there will always be a person who manages the process, who conducts this entire system and agents. So you don't believe in scenarios where artificial intelligence itself identifies a problem, proposes a solution, and writes the code itself. >> Look, there are reasons why we are engaged in responsible AI and why we test every model, why there are safety measures, vulnerability research teams, teams that try to hack the model, find out what system prompt is used. These are definitely things that we need to work on, just as we work on software security or traffic safety. Any technology we have in life carries risks and benefits. And I believe that when forecasting the future, we should focus on opportunities, and deal with risks every day so that they do not become a reality. >> Can you name your top three favorite AI applications? >> Well, of course, Copilot is my number one, because we are working on this project. >> Do you already use them within the team for writing code? >> Yes. Yes. Everyone at GitHub. Not just developers, everyone uses Copilot. Product managers, designers, HR, lawyers, financiers. Yes, everyone uses GitHub. That's the essence of GitHub. Everyone works at GitHub and uses Copilot. I really like ChatGPT for everyday tasks, various other questions. I have it installed on my Mac, and I think it's increasingly replacing regular internet search. And, um, what's my third favorite system? For example, I have an app on my phone to transcribe calls. I find it very useful, for example, for interviews. I think it's important to immediately warn the interlocutor that you are recording the conversation and that AI will make a summary. But there are things like this. I'm not particularly creative when it comes to drawing, so the ability to use one of these models to create an image for me, for a PowerPoint presentation, is also very convenient. Cool. Thank you for such a positive outlook on the future.