#270 Thomas Dohmke: GitHub CEO Reveals How AI Will Change Coding Forever

AGNTCY - Unlock agents at scale with an open Internet of Agents. Visit https://agntcy.org/ and add your support. In this episode, GitHub CEO Thomas Dohmke joins us for a deep dive into the evolution of software development — from decentralized version control to the rise of AI coding agents. With over 150 million developers on GitHub and tools like Copilot rewriting the rules of software engineering, we explore what it really means to build in an AI-native future. Thomas shares the origin story of Copilot, how GitHub is shifting from human-to-human to human-to-agent collaboration, and why he believes natural language is becoming the universal programming language. We also cover the technical architecture behind Coding Agents, the feedback loop between developers and AI, and what it takes to scale multi-agent systems in the real world. Like and subscribe for more! Stay Updated: Craig Smith on X:https://x.com/craigss Eye on A. I. on X: https://x.com/EyeOn_AI (00:00) The Future of AI-Powered Coding (02:04) Thomas Dohmke's Journey (05:16) GitHub's Origin Story & Evolution (08:45) Life Before GitHub: Early Version Control Systems (10:40) What is Git? And Why GitHub Matters (12:36) The Birth of GitHub Copilot (16:17) The Rise of AI Agents (17:52) How Kids Are Learning to Code with Copilot (22:38) Can Non-Coders Use Copilot Agents Effectively? (26:01) What the Coding Agent Actually Does Behind the Scenes (31:30) The Models Behind GitHub Copilot & Developer Choice (35:22) How Much Code Is Now Written by AI? (38:51) GitHub's Innovation Strategy (41:54) What's Next for GitHub (45:24) From 150M to 1B Developers: Empowering the World to Build (47:51) GitHub Universe & Galaxy Events (49:53) GitHub's Innovation Graph and the Power of Open Collaboration

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Speaker 00

  Views, coding agent, all you do is write the prompt. The prompt is a longer description, so arguably you're writing natural language to describe the problem you're trying to solve. And then agent mode or coding agent writes all the code. And so for certain scenarios, it's a hundred percent code written by AI. But you still have to write, you know, in, in, in English. So arguably, human language is becoming the universal programming language. In the future, the developer will have an orchestra of agents and the role will be to be the manager or the conductor of that orchestra. And I think ultimately the biggest skill for developers other than understanding technology and the craft of engineering will be to decide, am I doing it myself or am I using an agent?

• 0:41Speaker 00

  And at what point do I have diminishing returns to assign these three lines of code to an agent instead of just doing it myself?

• 0:48Speaker 01

  Building multi-agent software is hard. Agent to agent and agent to tool communication is still the wild west. How do you achieve accuracy and consistency in non-deterministic agentic apps? That's where agency comes in. A G N T C Y. The agency is an open source collective, building the internet of agents. And what's the internet of agents? It's a collaboration layer where AI agents can communicate, discover each other and work across frameworks.

• 1:29Speaker 01

  For developers, this means standardized agent discovery tools, seamless protocols for inter-agent communication, and modular components to compose and scale multi-agent workflows. Build with other engineers who care about high-quality, multi-agent software.

• 1:52Speaker 01

  Visit agntcy.org and add your support. That's A G N T C Y dot O R G.

• 2:03Speaker 00

  Hey, I'm Thomas and I'd like to say I'm not only a GitHub CEO, I also am a developer and I've started coding more than 30 years ago in the early 1990s. First on an East German clone of a Z80 and then on a Commodore 64 and later a Free 86 PC. I've been a developer since my mid-teens, I guess. And through different parts of my journey, I ultimately landed first at Microsoft who acquired my startup in late 2014 and then I was part of the team that acquired GitHub in 2018.

• 2:45Speaker 00

  Actually, you know, today, as we're recording, this is June 4, 2025, so we're exactly seven years after the announcement that Microsoft will acquire GitHub. And through that announcement and the acquisition, I joined GitHub and GitHub's leadership team in October 2018 and then became the CEO three and a half years ago.

• 3:06Speaker 01

  And so you were living where in East Germany before the wall came down?

• 3:12Speaker 00

  In Berlin. So I grew up in Berlin, you know, one of the suburbs that has all the prefab buildings called Mazan or Mazan in German. So I grew up there, you know, had a, you know, was 11 years old in 1989 when the wall, you know, technically fell, quote unquote. And then, you know, as Germany got unified in 1990, I finished high school in 97 and started a technical university in Berlin. In ninety eight and then moved down to Stuttgart to work for Mercedes. And so ever since I left Berlin in two thousand two and haven't been back other than for, you know, few business trips and to see the family.

• 3:56Craig S. Smith?

  And you you said you began coding on this East German clone. Was that before the wall came down?

• 4:02Speaker 00

  Yeah, I think my history is a little bit muddy on this. To my memory, we discovered in the geography lab these RoboTone computers, which were, you know, East German made computers, but very similar to Commodore sixty four. You would, you know, start with a basic prompt and then load. I think it was cassette tape in in in school. You would load software with a basic command. And so we started coding on that. And then the community center in the suburb had had these same computers. And we started going to computer club. I don't know when, I think that was after the wall had fallen. Obviously, the, you know, the suburbs and Berlin itself didn't change as rapidly as as it now feels in hindsight.

• 4:47Speaker 00

  And so a lot of things were still like following the East German system. I think the education system only switched in September ninety ninety one, given that the unification day was October, which was after the school, which was after the school had started. So there was another year from ninety ninety to ninety ninety one, where we were still in the East German education system, even though technically the country had already been unified. And then only in ninety ninety one, we switched to the best German system.

• 5:16Craig S. Smith?

  So GitHub, I mean, I think of GitHub still, and I'm not a coder, as a code repository, but it's become much more than that. Can you give us just a thumbnail history of GitHub's development? And then we'll talk about some of the new things that you guys are doing.

• 5:34Speaker 00

  I would argue GitHub was always more than just a repository. The repository is the foundation, you know, the primitive, as we would say, that has enabled, you know, what you might now call modern software developer collaboration. But it essentially invented how developers work with each other. And in October two thousand seven, I think, is when the founders of GitHub, Chris, PJ and Tom, started working on it. A few years ago, we celebrated that by creating a poster that had the first ten or so. Commit messages on it and as you may imagine, a lot of those were kind of funny because obviously

• 6:16Speaker 00

  the founders at that time didn't realize what they will be building and how important that will become for the world. It officially launched early 2008, depending on whether you count, you know, the preview or the actual release somewhere between February and April. A lot of the early users of GitHub were in the Ruby on Rails community, given that the GitHub founders were coming out of the community in San Francisco. And it started really as repositories. The pull request wasn't actually there in the beginning, but it came very, very quickly. But it already had like, you know, it's hard

• 6:59Speaker 00

  to think about this now as something revolutionary, but in 2008, being able to see a repository in a browser web interface, clicking on it was actually fast. It had a nice, modern user interface. That alone, I think, created a lot of attraction for folks to say, I host my repo on GitHub. It's easy to set it up. It's easy to push my code and then I can just go there and click through the file tree and see the history of the files and the commit messages and so on. So repositories was that was that first building block collaboration with pull requests that GitHub ultimately invented was another one before the pull

• 7:40Speaker 00

  request. Open source maintainers would send this, you know, files with changes around and then apply those changes locally to merge them back into the into the main repository. And later came issues and wikis and all these things that developers need to collaborate. And today, I think what defines GitHub is that whether you're an open source developer or a commercial developer, whether you're working on a startup or a hobby project, all that software development process is very similar to each other. While we think about these as different worlds, they're actually not that different. And oftentimes, you know, the four things I described are all the same person. Yeah. And just in different times of the week, you know, they're doing open source, they're

• 8:24Speaker 00

  doing hobby projects, they're working maybe on a big idea, and then they they also have a job where they're using GitHub.

• 8:29Craig S. Smith?

  You know, it's always fun to look at people's profiles and all the the different things they're working on. What, what, just for, for the kids out there. What, what, where did people put their code before GitHub?

• 8:45Speaker 00

  That's a, you know, there's many ways to answer that question. Um, and the, the funniest, I think, is that you would just put it in a folder on your computer. And then you have a V1, V2, or, you know, what, you know, date and, uh, uh, in time, uh, stamp. And then you would realize three days later that, um, you didn't actually do that. And then you lost a lot of the changes or you can't roll back to what you wanted to do. And, um, I actually think this is even today is still pretty prevalent that people just do that. Um, um, if they're early in Korea, if, uh, you know, if they're working on just some, uh, of, if their first steps in software development, uh, a lot of the code is just stored locally.

• 9:28Speaker 00

  Um, the origins of version history is, uh, version control systems is probably like a history podcast in itself. Um, there was something called CBS, um, uh, which was, um, the most used, uh, version control system. And I, uh, uh, started coding and, and during university and then subversion, uh, came out and that was, uh, very popular. And, um, SourceForge was, was kind of like the place where you put your, uh, subversion repository to share with the world. And then Git was developed by the Linux kernel team, Nostorwals, uh, and others. And Mercurial was the competitor, um, uh, at the time and, and clearly Git, you know, has, has won that race. But in, uh, 2008, when GitHub, uh, launched, um, that, that was still a question.

• 10:13Speaker 00

  Is it, is it Git or is it Mercurial? But a lot of the early GitHub users came from SourceForge and, um, uh, from Subversion. And even today, we're still migrating customers of their old Subversion repositories into, into Git and GitHub.

• 10:28Craig S. Smith?

  Yeah, uh, yeah, for, again, for listeners that aren't, uh, coders, uh, describe what a Git is and, and what.

• 10:37Craig S. Smith?

  GitHub did for Git.

• 10:40Speaker 00

  In its most simplest form, I think the way to, to, you know, look at Git is that it stores changes to the text files that, that you're writing and developers, obviously, you know, as they're writing programming language, they're writing files, you know, plain text files with the code in it and Git stores the changes that you make to that file at the time when you commit into the repository.

• 11:02Craig S. Smith?

  Yeah.

• 11:03Speaker 00

  And so this version control, it's the history of all the changes and nowadays, you know, in Word and in Google Docs and elsewhere, you can also go to the Word history, but that wasn't always the case. And the, the significant change of Git compared to other version control systems that came before it is that you have all the repository on your local machine and it's a decentralized technology. So you can have a copy of the whole repository with all the, all the branches and all the changes. I can have my own local copy and GitHub offers us a place where we can bring those together and where I can push my changes and you can pull my changes and we can both collaborate on the same repository with each other.

• 11:46Speaker 00

  And so GitHub, as the name implies, offered the central place for all these decentralized repos, so we can share them with each other because as much as we, you know, like decentralized technology and like to talk about this decentralized world, the reality is you're the place, a home, if you will, to go to. And, you know, we're calling GitHub the home of all developers because we believe that that's, that's our aspiration, if you will, to build a home where every developer goes and, and likes to collaborate with others and meet with them and exchange code.

• 12:23Craig S. Smith?

  And so let's talk about some of the new features that you've developed, particularly it's been, you've been very active this year. What, what is top of the list for you?

• 12:35Speaker 00

  AI clearly is top of the list. Otherwise I wouldn't be on this podcast, I think. And it's interesting, because we talked about Git and GitHub. And Git is a technology that wasn't invented by the GitHub founders. It was invented by the Linux kernel team. And then the founders took that a few years, I think, two years before GitHub. And then the founders took that technology and built GitHub. And the same happened to us again with Copilot, where a technology was invented by Google, you know, the transformer model, and then taken by OpenAI to build back in 2020 G P T three and a version of that model fine tuned called Codex, fine tuned on open source code.

• 13:20Speaker 00

  And we at GitHub at the time, you know, got access to this model through the partnership between Microsoft and OpenAI and start playing with it and asked questions, you know, how write a method that detects prime numbers or sort, you know, an array, typical kind of like low, lower entry, entry level coding questions for, for students in university. And it was able to write these, these code examples in, in proper programming language. And in fact, it could do that in different languages and would not mix things up between, you know, some languages like have parentheses, other have curly braces, some have commas, other have, you know, colons and whatnot.

• 14:04Speaker 00

  And that, you know, led us to the, you know, idea of building a product around this, which, which then became Copilot. And in its earliest form, it was really simple. It was just auto completion in the editor. So where developers write their code and where they're using Git, you know, to, to create the version history of that code, they would now get a suggestion from, from this Codex model or in general, the large language model. And the suggestions wasn't just like the next word or the next few words. It was multiple lines of code, you know. A lot of developers, a lot of developers do is write in test cases and writing what we call boilerplate code, things that you repeat over and over again, you know, making rounds in corners.

• 14:49Speaker 00

  So your user interface looks nice and so on. And so predict that code in a such a kept developers in the flow, whilst instead of switching between your editor and the browser or the documentation or the editor and asking, you know, somebody in your office or when you're guest in a on Slack or Teams, Copal gave you enough of an idea of what you could do next. Even with hallucinations and all that, then you could just keep staying in the flow. And often this flow state feels really magical for software developers is kind of this idea. Now, all I need to do, code, is just to, you know, take this idea and write down all the code

• 15:31Speaker 00

  to build what I have in my head. And so five years later, we're still working on that. And you know, there's still a lot of innovation happening in this code completion space of predicting what's the next change after the current change, what's the next change in JSON tabs or just making making the model better. But ultimately where we're heading into is agents and building agents that developers can use to write code, we call that the coding agent, to review code, we call that the code review agent, or to fix their security vulnerabilities, and we call that the auto fix agent.

• 16:07Craig S. Smith?

  Are all of these under the GitHub coding agent umbrella? Or is the GitHub coding agent referred to one of those?

• 16:17Speaker 00

  It's all under the GitHub Copal umbrella. So you know, the way to think about Copal is has all these capabilities and we're adding more. So just like a human developer, you know, the Copilot, as a developer learns more skills and gets better in each of them. And so Copal has a coding agent, a code review agent, an auto fix agent, it has auto completions, it still has chat.

• 16:39Speaker 00

  Which is often quite useful. You know, I have two sons, they're both coding in Python and so they're using Copilot to fix their own bugs instead of, you know, coming into my office and asking me. And often chat is the better way for them to learn than just having the agent do it because they understand then, you know, from the description and the code example of how to make the change instead of just magic happening to them.

• 17:05Craig S. Smith?

  Yeah, actually, how old are your sons? And did you teach them to code or have they learned on their own or at school?

• 17:14Speaker 00

  They are thirteen and ten. And it's a mix of all these things that you said. So we encouraged, you know, them to learn with, especially, you know, during the pandemic.

• 17:27Craig S. Smith?

  Yeah.

• 17:29Speaker 00

  They did, I think typing was the tool they used to learn, ten finger typing. I think they can do it better than I actually because I never properly learned it. And so it's more like eight finger typing or seven finger typing. And they use Scratch from MIT and Lego Mindstorms, which has a very similar interface as Scratch, like, you know, logic blocks and whatnot. They learned Python in school, although the eldest learned Python in middle school. And then the rest, you know, they taught themselves very similar to how I taught it myself in the 1990s, except, you know, I only had books and magazines and they have the whole range of the internet.

• 18:10Craig S. Smith?

  Yeah.

• 18:11Speaker 00

  And that's incredibly powerful, you know, even without AI, just go on to, you have a problem, you know, you're trying to figure out how to do, I don't know, animated sprites in Python and Pygame. And you can just find a YouTube video for that.

• 18:24Craig S. Smith?

  Yeah.

• 18:24Speaker 00

  In fact, you know, my kids also taught themselves, you know, to solve the Rubik's Cube by just watching YouTube videos, right?

• 18:31Craig S. Smith?

  Yeah.

• 18:31Speaker 00

  And of course, there's a lot of fine error, but that is very similar to how many of us learn coding. And then Copilot gives you this, this, you know, extra.

• 18:41Speaker 00

  Extra step where it can look at your code, explain it to you. Not only in English, but actually also in German and any other languages. Which my kids speak both languages, given that they immigrated in the United States more than 10 years ago. But even that, I think, is incredibly democratizing. That as a seven, eight-year-old, if you want to learn coding, you can just ask it in your mother tongue. And it will explain to you, okay, you need to open a file and then this way you put the Python code. And this is how you run it. And now you have a snake game.

• 19:16Speaker 00

  I think kids are in many ways actually better at this than adults. Because it's so natural to them to just keep trying, right? They just keep trying. They just keep asking. As a parent, you quickly run out of patience or daytime. But kids will just keep asking Copilot or ChatGPT or any other AI tool. And they will not sit there and say, this is bad. Or I could do that way better if without the model. No, no, they will take the feedback and work with whatever Copilot gave them to make that thing work.

• 19:55Speaker 01

  Building multi-agent software is hard. Agent to agent and agent to tool communication is still the wild west. How do you achieve accuracy and consistency in non-deterministic agentic apps? That's where agency comes in. A G N T C Y. The agency is an open source collective, building the Internet of Agents. And what's the Internet of Agents? It's a collaboration layer where AI agents can communicate, discover each other, and work across frameworks.

• 20:37Speaker 01

  For developers, this means standardized agent discovery tools, seamless protocols for inter-agent communication, and modular components to compose and scale multi-agent workflows. Build with other engineers who care about high-quality multi-agent software.

• 20:59Speaker 01

  Visit agntcy.org and add your support. That's A G M T C Y dot O R G.

• 21:11Craig S. Smith?

  Yeah, a couple of questions about the Copilot Coding Agent. For, I mean, it's a tool built for coders. But for non-coders coming into this world, is it reliable enough that a non-coder could open up the Copilot Coding Agent, ask it to begin coding something with chat? And because, as I said, I'm not a coder myself. When I, when, what was it called back then?

• 21:53Craig S. Smith?

  You know, ancient histories, like three years ago. The first of these that came out, I can't even remember what they were called. But, but, you know, you, you ask it, ask it to code something. You then copy it into, you know, Visual Studio or whatever. You have, you run it, you get an error, you cop the error back. And you go back and you just end up going down a wormhole and never get it resolved. So is, I mean, where are we now with these agents? Is it to the point that a non-coder could start using them?

• 22:38Speaker 00

  Yeah, I mean, there wasn't even Visual Studio when I started. Basic on the PC, it was Turbo Pascal. Turbo Pascal that actually ran in DOS and on Windows. And then to run the application, you would, you would bring, you would move Turbo Pascal or on Unix back then it was VI, would move that into the background. You know, type the comment line command to run the thing or compile the thing. And you run it, you get an error message, you have to figure out, okay, so how do I bring back the error? And you wouldn't even see the error message in the editor side by side. Those things were in separate, on separate planes, if you will. Although there wasn't Windows, right?

• 23:23Speaker 00

  Like it was all.

• 23:26Speaker 00

  Today, I think the state of the art is, and the answer to your question is both. So as a non, somebody who doesn't understand programming language, you can use tools like Copilot.

• 23:42Speaker 00

  You know, maybe I should explain what the coding agent is, but you can use tools like Copilot. And there's a bunch of others out there, like Russell has V Zero and Lovable Bolt, some of those. You know, do similar things, which is they can, Net open AI has Codex now, can write a prompt. And then it, you know, creates all the code for you. And especially in web development, like create a quick webpage or a small web application. Even a little game, as long as it runs in the browser, you can easily create with these things. And you can keep going with more prompts in the same way that you can keep going.

• 24:23Speaker 00

  You know, when you wanna render an image in ChatGPT and you don't like the first version of that image, you keep going writing more prompts. And eventually you either get there or like, okay, so I got so far off what I had in my head that it's easier to start from scratch. Yeah. And I think that's between an image model or an image, you know, generator and a code generator, it's very similar. But you can get a long way, but if you don't have knowledge of the programming language and, and you know, how this all works behind the scenes. You're going to get into a place where at some point a prompt doesn't get you where you need to go. Right. And certainly when we think about, you know, scaling this then to, you know, millions or hundreds of millions of users.

• 25:05Speaker 00

  At some point, you certainly have to go and, and, and look, become a professional developer or hire those. Yeah. And the other piece is that, you know, and often this is called vibe coding, right? You're vibing by creating this, but you're always, you know, I think going to reach the point where now you have to learn what actually was generated by, by, by Copilot. And to understand whether that code is actually, you know, doing what you described, you wanted to do. That's even more true in, in professional environments, right? Like. When you join a company that builds software, in ninety-nine percent of the cases, you're taking over somebody else's code.

• 25:50Craig S. Smith?

  Yeah.

• 25:51Speaker 00

  And you're working, or you're working on a project with, you know, that had dozens, if not hundreds, of developers already on it. And so working on an existing code base, and that's where the coding agent comes in. The idea of the coding agent is that you describe the changes that you want to make to the existing code base. And you sign Copilot to it, and then Copilot, in the background, AKA in the cloud, spins up a virtual machine that checks out the code, you know, it installs all the dependencies and all the tools it needs. And then it tries to figure out with the help of a model how it implements the issue or the task that you described. And as it does this, it submits code into a pull request, just like a human developer would.

• 26:32Speaker 00

  And it describes, you know, how it made these changes and then how it, you know, went through its plan. But then you need to come in and say, okay, this is actually what I wanted. And this fits, you know, into the coding practices of my team, and it uses the right database and the right cloud provider, and it doesn't introduce any new security vulnerabilities. And that's where a professional software developer will always have to understand the code that the agent generates within the, you know, within the project they're working in.

• 27:03Craig S. Smith?

  Yeah.

• 27:03Speaker 00

  So you're sitting on the spectrum of, if you want to do something without coding, just to spin up, you know, your wedding webpage, or like a quick e-commerce shop connected to Shopify, you know, to sell something on Black Friday, that definitely will work. But there is a, on that spectrum, there is a point where you have to then also understand, you know, the concepts of computer science, computer engineering, to ultimately be a professional software developer, and that profession isn't going anywhere.

• 27:34Craig S. Smith?

  Yeah.

• 27:35Speaker 00

  It will continue to exist, and it's going up. It's not going away, is I guess what I'm trying to say.

• 27:41Craig S. Smith?

  Yeah, yeah. And what you described, a software developer coming into work on an existing code base.

• 27:51Craig S. Smith?

  Does Copilot Coding Agent, how much of the codebase does it review before it starts

• 28:02Craig S. Smith?

  making suggestions or operating in a virtual machine in the cloud? Because, you know, there are so many, you change something, it could have implications that you don't see in your, in the amount of code on your screen. Yeah, so does, does it review the entire codebase? Or do you choose how much for it to see? Or how does that work?

• 28:33Speaker 00

  That's actually fascinating, you know, to, to observe for different types of issues assigned to the coding agent. So it works, you know, in many ways, like a human developer would work in the codebase. So it checks out the codebase, you know? It reads, you know, the description of what it needs to do. And then it feeds that into the model to generate the chain of thought. And as it does that, the model is allowed to call tools. And these tools can, you know, be things like using code search to find, you know, code within files, to identify the files it needs to modify. It can be tools that it runs on the command line, you know, to install something

• 29:18Speaker 00

  or to update a package or maybe even add a package. And what it does, that comes out of the chain of thought. That comes out of the model as it describes, you know, how it would solve that problem. And, but the key here is that it then looks at the output of these tool calls and, you know, trying to compile the thing or it can use a technology called MCP Model Context Protocol that was invented by Anthropic, that lets it effectively connect to every, every other system that already has an MCP server, that's one for GitHub and one for Figma, which is a design tool. And there's even one for, you know, can have it connect to your Apple Notes or whatnot on your local machine.

• 30:01Speaker 00

  And so it effectively works like a human developer tries to figure out within the codebase how to do the task. And then it iterates on that with the help of reasoning and the chain of thought. And until it ultimately gets to the point where what it has generated, compiles and, you know, fulfills the test cases, it created itself and then it submits that for code review. But obviously, you know, that's never going to be perfect, right? There's always going to be either it thinks it has succeeded and you look at it like, what is this? And then you can give feedback. So we believe there's always going to be a feedback loop between the human developer that has signed the task to the coding agent and the agent itself.

• 30:46Speaker 00

  And that feedback loop at GitHub lives in the pull request, just like the human collaboration between two developers lives also in the pull request. So if you go back in 2008, GitHub was human to human collaboration in software projects. Now in 2025, GitHub is still human to human collaboration, but we are extending it to human to agent collaboration.

• 31:09Craig S. Smith?

  Yeah, and these models are getting more and more powerful.

• 31:17Craig S. Smith?

  The the the underlying model for GitHub code pilot coding agent is, are the OpenAI models, is that right?

• 31:30Speaker 00

  For the coding agent today, it is the Claude Sonnet four model that just came out. For Copilot itself, the Copilot that you use, you know, in your IDE, it is a range of models. Code completions or autocomplete is on an OpenAI model that's based on the four mini. We also have, you know, for chat, we have four one and four five and Claude and Google models. And then for agent mode, it's those models that support, you know, have reasoning capabilities and can call tools, you know. At the end of the day, whether it's the open source projects you're picking on GitHub,

• 32:14Speaker 00

  or whether it's the models we believe in developer choice. A developer tool is not going to be successful if it doesn't offer developers the choice between the different options that are available in the market, and let them pick, you know, which is the best model for their scenario. And what best means, and I'm making air quotes for those that are listening audio, what that means is highly dependent on the developer themselves, right? They might have previous experience, they might have a team that says, "We're only using this model or this open source library." They might have certain beliefs, and all these things play a role, and we are not the ones judging that. We are the ones offering that choice. And that goes back to what I said earlier.

• 32:59Speaker 00

  We want to be the home of all developers around the world, and as such, we believe in that choice.

• 33:04Craig S. Smith?

  So the developers can switch between models. Presumably, you know, you can try it with, you know, doing, writing a piece of code with one model, and then try it with another model and see which one you like better. But is there an orchestration layer where

• 33:25Craig S. Smith?

  the Copilot Coding Agent makes that decision for you?

• 33:31Craig S. Smith?

  You know, it decides on its own which is the best model to use?

• 33:35Speaker 00

  Today, the Coding Agent is only on Sonar four. But that's certainly something that we have in our minds or in the backlog of how we say in software development,

• 33:48Speaker 00

  to have more orchestration on these things. And man zu bringen in the Code Review Agent, which runs on a different model and reviews the code in the pull request or the AutoFix Agent that fixes security vulnerabilities. We are already supporting a range of models, and in some cases, we pick the model. Like in code completion in this agent scenario and in others we give developers the choice. I think the crucial thing is there is no single best model that works for every single developer out there, right? There's many programming languages, there's many frameworks, there's many styles of how software is developed in companies. And we believe developers will figure out what works best for them.

• 34:33Speaker 00

  And often that is trial and error, just as you do trial and error in other parts of the software development lifecycle. And you know, we all have figured out at some point that we made a wrong decision on some framework library. A decision in life, I guess. And the nice thing about software and models is that it's really easy and cheap to switch to another model and then try again. But yeah, this auto orchestration, I think, will play an increasing role for those that do not care what model they want to use. And they just want to, you know, have the AI decide for them and offload that decision. And so it's going to be mixed off of those two things.

• 35:15Speaker 00

  Those that want choice will get choice and those that want the default will get some form of auto mode.

• 35:22Craig S. Smith?

  Do you have a sense of how much of today's code is being written on GitHub?

• 35:29Speaker 00

  Oh, how much code is written on GitHub? I don't have that number. It's a

• 35:33Craig S. Smith?

  Not how much, what percentage of new code in the world is written on GitHub?

• 35:39Speaker 00

  I have, I have, I have no idea of lines of codes. I mean, we have 150 million developers. I can look it up. We publish every year what we call the octaver on statistics from open source. Obviously, we don't want to talk about, you know, what enterprises do in their private repositories. But if you just look at pull requests, we're talking in the range of hundreds of millions of pull requests every year. And so then, you know, if you do the math, if each of those pull requests only has like, you know, 10 lines of code, that would be already a billion lines of code. I'm sure the number, the number is way bigger, especially as, you know,

• 36:22Speaker 00

  Developers change existing code all the time, right? Like, they either do that intentionally by refactoring something like changing a long method with lots of lines of code into a smaller method because at that point where they were working in it, they realized the previous developer just got it done and now it's time to clean up.

• 36:41Craig S. Smith?

  Yeah.

• 36:42Speaker 00

  Or because they're using something like agent mode and they're asking the agent to make a change and then the agent decides, okay, you know, as part of that change, I'm also, you know, changing some other files and make those better. And so, you know, if you look at just AI, it's been over two years ago, I think in early 2023 is when we said Copal is writing forty-six percent of those lines of code in, writing forty-six percent of lines of code in those files where Copal is enabled.

• 37:20Craig S. Smith?

  I see.

• 37:21Speaker 00

  It was already half of the code through just auto-completion, half of the code more than two years ago. And today, if you use agent mode in VS Code or if you use coding agent, all you do is write the prompt. The prompt is a longer description, so arguably you're writing natural language to describe the problem you're trying to solve. And then agent mode or coding agent writes all the code. And so for certain scenarios, it's a hundred percent code written by AI, but you still have to write, you know, in English or whatnot, what you want the agent to do. So while you're longer writing programming language and code, now you're writing a specification.

• 38:06Speaker 00

  And so arguably, you know, human language is becoming the universal programming language. And that's still written by the person that has the idea in their head and tries to take that idea and transform it into something real.

• 38:19Craig S. Smith?

  There are other DevOps platforms or whatever you want to call it. GitLab is one that comes to mind. I mean, GitHub is the dominant, but how? Many other platforms are there that you're watching? And do you have to? I mean, obviously, you have to pay attention to the competition, but there's such a critical mass behind GitHub today that it'll take a long time before anyone challenges your position.

• 38:51Speaker 00

  I think it's in our nature to always be worried about, you know, even if that that's probably true what you said, that it's going to be a long time before a large platform like GitHub is going to be challenged in its position in terms of, you know, losing all its value. You know, there's many other technologies in software and in the personal computer world out there that are old and still exist. You know, I saw over the weekend some YouTube video where they were talking about some industries still hiring Windows 95 developers because they're running such old. I think it was air traffic controllers, actually.

• 39:34Speaker 00

  I think it was air traffic controllers that are still running on super old versions of Windows in some towers. And so technology never goes away. But I think as Microsoft, as part of Microsoft and as GitHub, we always have, you know, a healthy level of anxiety. I'd say of what else is happening around us? Are we going to be disrupted? The innovators dilemma, you know, the Hayden Christensen book plays a big role, I think, for most companies that have research of certain size. And as such, you know, whether it's in DevOps, whether it's in generative AI for software developers, whether it's in security for software developers,

• 40:18Speaker 00

  you know, with our partners at Microsoft, we also, you know, own two of the largest IDEs, largest editors, Visual Studio, Visual Studio Code. We have the .NET ecosystem. We are part, we're the owners and stewards of NPM, the Node Package Manager, which is the largest package registry for everything in JavaScript, right? So. What I'm trying to say is, you know, GitHub is GitHub and we're going to keep innovating. We're going to keep focusing on what developers expect from us. While also, you know, looking at what's going to be ahead. Where do we need to innovate?

• 40:59Speaker 00

  Not only to grow our business, but ultimately to prevent that we're going to be disrupted by others. And I think the great thing about GitHub and being part of GitHub is that everybody at GitHub is using GitHub. Not only the engineers and the product managers, but also, you know, the leadership team, the human resources department or the people team and the legal team. And they have, you know, their terms, our terms of service are just a GitHub repository. Because at the end of the day, everything is a pull request against, you know, some previous version of that same file or adding a few new files to the repository.

• 41:39Speaker 00

  So I think as a developer myself, building tools for developers is, you know, my dream job. And we're going to keep pushing on the GitHub side.

• 41:50Craig S. Smith?

  Can you talk a little bit about what's next?

• 41:53Speaker 00

  It's agents. You know, we already mentioned the coding agent.

• 41:58Speaker 00

  If you look into the space of these agents right now, there's a very popular benchmark called SW bench, S W E bench. And the best models in the SW bench are somewhere in the 60% range. Maybe it's 70% by the time, you know, this comes out. But that's only, you know, about 2,000 issue pull request pairs out of a dozen Python repositories. And 70% is not a lot. Because what's for the remaining 30%? And what's with all the other programming languages? And the team behind SW which actually recently expanded into multiple programming languages.

• 42:40Speaker 00

  And then the best models are much, much lower, I think in the thirty percent range. So there's a lot of work to do to get these coding agents, as exciting as they are and as promising they can be for developers to offload some of the work they don't want to do to just assign it to the agent. There's a lot of work to do to get this to a point where we can say, you know, mission accomplished and agents can now write all my test cases, write documentation, you know, fix security vulnerabilities, all the things that developers have to do in addition to build innovation.

• 43:16Speaker 00

  There's many other agents that we have in mind. We already mentioned the code review agent, the auto fix agent. At Microsoft Build, we announced the SIE agent that monitors your servers and then takes action, you know, based on findings, whether it's, you know, errors that are piling up or server load going high and things like that. And so you can imagine, you know, in the future, the developer will have an orchestra of agents and the role will be to be the manager or the conductor of that orchestra. And I think ultimately, the biggest skill for developers other than understanding technology

• 43:56Speaker 00

  and the craft of engineering will be to decide, am I doing it myself or am I using an agent? And at what point do I have diminishing returns to assign these three lines of code to an agent instead of just doing it myself?

• 44:10Craig S. Smith?

  Yeah.

• 44:10Speaker 00

  And predicting that and how good does my description have to be? And how far do I have to break the problem into small chunks so the agent can pick it up? That's going to be the job of the human and I think that's actually exciting because, you know, when I have an idea that I want to build something as a software developer myself. The hardest part is not coming up with this idea. The hardest part is to take this idea and convert it from the language I speak in my head, the English and German into a coding language. And part of that journey is like breaking into smaller and smaller chunks until you get to the point where you're like, "Oh, okay, so button, I know how to do that." You know, put a button on a webpage and then you go into look in the documentation of this.

• 44:56Speaker 00

  Now how do I handle that? You click on the button. That's the level where, you know, a few years ago, and we're going to go higher in this abstraction ladder, but we're still going to have this, you know, systems thinking process, the system design process of taking my idea and breaking it down into chunks that I can assign to the agent.

• 45:15Craig S. Smith?

  Yeah, yeah, it's fascinating. I was asking earlier about percentages. What percentage of the world's developers do you think are on GitHub?

• 45:24Speaker 00

  Well, 150 million, I think, you know. If you ask, if you ask some of the analysts out there, they're going to tell you that number is already higher than the number of professional developers that we estimate out there.

• 45:35Craig S. Smith?

  Yeah.

• 45:35Speaker 00

  And the question really is what counts as a software developer? Is a, you know, an eight-year-old that in middle school, what is it, not, not, who are you, 11 years old, I guess? A sixth grader in middle school when they do, you know, learn Python or they're working on robots, you know, for robot fighting league and things like that. Are they a developer or are they, you know, a student learning to code? The same way that they're learning physics and math and all that. And I think you wouldn't call, you know, a middle schooler a physicist just because they had, you know, two years of physics in school. But I would, I would hope, you know, they're, they're, they're on GitHub

• 46:19Speaker 00

  if only, you know, to learn from, from other developers out there. That, that really, I think truly is the magic behind open source that the majority of software developers on this planet have decided to share, you know, their work, you know.

• 46:34Speaker 00

  Sometimes it's just a short code snippet, right? And what we call a gist on GitHub. They share their work with the world, so others can learn from it, can, you know, take it, fork it, remix it. And we believe in, you know, a part of our vision is that we get from 150 million developers to one billion developers. Mostly because one billion, not because one billion is a round number, but because we believe that everyone on this planet should be able to create software if they'd like to do that. And that means they all have to learn it in school. I think computer science should be standard, just like math is a standard in school.

• 47:17Speaker 00

  And they might be vibe coding it, or how you said earlier, they might just build something without understanding a lot of the code. But they should be empowered to build software and not only consume software. Today, we are very much in that side of, we're all consumers on our smartphones. And I think we should all become builders if we'd like to, if we'd like to be a builder.

• 47:38Craig S. Smith?

  And you have events, is it GitHub Universe and GitHub Galaxy? Can you tell us what those events are? Are those important for community building?

• 47:49Speaker 00

  GitHub Universe is our yearly conference. It's traditionally in San Francisco, sometime in either late October or early November. This year, it's October 28, 29, I think. I might be off by a day or so.

• 48:06Speaker 00

  And it's a big gathering of software developers that work, you know, with us or have their code on GitHub or want to learn more about Copilot and AI. And it's in Fort Mason in San Francisco. Oftentimes, it's the time of the year when San Francisco is actually nice weather. It's sunny and blue sky. And it's important for us, you know, from an announcement perspective, but it's also important because we get to meet many other developers from around the world. And it's a gathering, if you will, you know, to share and collaborate and...

• 48:45Speaker 00

  Galaxy ist dieses Jahr ein virtuell ein Event, wenn es mehr ums How to use GitHub ist, es ist wie ein Lernen über GitHub im professionellen Umfeld. Wir sind Teil vieler anderer Events, die Microsoft built, die größte Developer Conference, und du findest Hubbers, wie wir einfach Kollege, wir finden Hubbers an vielen Events, ob wir uns ein Booth oder eine Session haben. Ja, wir oft anbieten, dass wir auf unseren Social Media Accounts folgen, folgen GitHub on X und anderen Plattformen, um mehr zu lernen, wenn Sachen kommen.

• 49:28Craig S. Smith?

  Ich wollte etwas über eine der Dinge, die du herausgefunden hast, recently ist Innovation Graph Data. Kannst du etwas darüber sprechen? Ist das, und wo ist das in deiner Sicht? Ist das eine wichtige Werkzeug oder eine Verfügung? Es klingt für mich enttäuschend.

• 49:53Speaker 00

  Es ist wichtig, weil wir glauben, dass es Teil dessen ist, was wir tun, um der Community zu helfen. Und GitHub hostet all dieses Open Source Code, und das ist natürlich nicht nur interessant für diejenigen, die mit dem Code arbeiten und in ihrem eigenen Projekt eingebettet sind. Es ist auch interessant für Forscher, Politiker, um besser zu verstehen, wie die Welt innoviert auf GitHub. Es bringt uns um den Kreis, die Gespräche, die wir im Anfang des Raums hatten, die ist, GitHub ist das Land, wo...

• 50:36Speaker 00

  Millions of developers around the world collaborate with each other. You know, we once called it the largest team sport on earth. There's not many things on this planet where people from any, almost any country, from any background, often you don't even know their names, certainly you're not looking them up in some, you know, organizational chart and look up what's the title and the background and whatnot, right? What you look at is, okay, is the code or the issue or the asset, whatever the person is contributing to my project, is that useful for me? Is that aligned with where I want to go with my open source project?

• 51:19Speaker 00

  Maybe I tell them, hey, this is cool, but like, I recommend you fork off and, and so it's important for us that, you know, people around the world that are outside of technology or that are adjacent to it by researching the social dynamics behind it, that they have a tool like the Innovation Graph and the Archive Program and many other things that we did over the last few, over the last decade,

• 51:47Speaker 00

  that they have access to that information beyond just the source code.

• 51:51Craig S. Smith?

  Is there anything I didn't ask that you'd want to listeners to hear?

• 51:56Speaker 00

  I hope listeners, you know, took away from this conversation that A, we are really excited about, you know, the future of software development and we don't believe that AI will replace developers. We in fact believe AI will empower even more developers on this planet. Those that just want to do that because I want to automate something on their computer, you know, taxes come to mind and all these other chores that you have to do, even though somebody promised to you that once you get a personal computer, everything gets easier. Or those that want to become, you know, the next startup founder, the next professional developer in a big tech company or, you know, the next GitHub employee, the next Hubber

• 52:41Speaker 00

  that works on GitHub and other developer tools to contribute back to the community. So we're really excited about that world and we think AI is going to accelerate the way software is developed.