No Priors Ep 106 | With GitHub CEO Thomas Dohmke

(instrumental music) . Hi, listeners, and welcome back to No Priors. Today, we're joined by Thomas Dunk, the CEO of GitHub, a platform used by over 150 million developers worldwide to collaborate and build software. As CEO, Thomas has overseen the development of tools like GitHub Copilot. Before becoming CEO, he helped shape GitHub's product strategy and power its global expansion, and previously worked at Microsoft. In this episode, we'll talk about the future of software development, the role of AI in coding, open source, and product plans for Copilot. Thomas, welcome to No Priors. Maybe we can start with the meat of it. What is happening with, um, Copilot and, uh, the new releases at GitHub recently?

You're heading straight, straight into it. Um, we're really excited about, you know, making Copilot more agentic. A few days ago, we announced the agent mode, uh, in, in Copilot and VS Code. Um, so instead of just, you know, chatting with Copilot and getting responses and then copy and pasting the code e- into the editor, or, or using autocompletion, the original Copilot feature, you can now, uh, work with an agent and it helps you, you know, to implement a feature. And when it needs to install, like, a package, it, it shows you the command line, terminal, commando, and you can just say, "Okay, run this." Um, you're still in charge, right? So that's the k- the crucial part of, um, uh, these agents that we have available today. That as the human you're still, uh, as the human developer, you still need to be in the loop. Uh, but we also showed, you know, uh, a teaser of what's about to come in, in 2025. Um, we call this, uh, Project Padawan, you know, because it's like a Jedi in a Padawan. You, you gotta have patience and you gotta, you know, learn how to use The Force. Um, but we think, you know-

(laughs)

... uh, in 2025, we get into a place where you can assign a GitHub issue, a well-defined GitHub issue, um, to Copilot, and then it starts creating a draft pull request and it outlines the plan and then it works through its plan. And you can, similar to how you observe, uh, a coworker, like you can see how it commits changes into the pull request and you r- can review this and, uh, and provide feedback t- to Copilot. And so, it, uh, Copilot b- basically graduates from a pair programmer to a peer programmer that, that becomes a member of your team.

The obstacles to that right now are some new model advancements. Is it just building out some other core technology? Is it just the UI? Like what, what is keeping that from happening right now?

Yeah, I think the first thing is the model, uh, the full o3 model that's not available yet, but OpenAI showed, um, uh, as part of the Shipmas, uh, uh, right before the holidays. We're going to see, you know, improved reasoning. And I think it's as the models get better in reasoning, um, we're going to get closer to 100% of this V-Bench, which is that benchmark, uh, out of 12 repos, um, uh, open source Python repos. Um, a team in Princeton identified, uh, 2,200 or so, uh, issue pull request pairs. Effectively all the models and agents are measured against. And so that's number one, you know, the, the model and the agent combination. I think the second piece is just the figuring out what's the right user interface flow. Um, if you think about the workflow of a developer, right? You f- you have an issue that somebody else filed for you, you know, user, quarker, product manager, or something that you filed yourself. Now how do you know whether you should assign Copilot to this, um, the agent, uh, to it, um, or, or whether you sh- need to refine the issue to be more specific, right? It's, it's crucial that the agent is predictable. That you know that this is a, a task the, the agent can solve. If not, then you need to steer it. So steerability is the next thing, and to either, you know, extend the, the definition, um, uh, or the agent needs to come back to you and, and ask you additional questions. And then at the end of the process, you wanna verify the outcome and, and so in our demo, that's where we're thinking the right flow here is actually that the agent works in a pull request, like similar to a human developer with loads of commits and then you can roll back those commits or, or check them out in, in, in your VS Code. We saw that with, with some of the agents that are available is that, do I, as a developer actually tolerate (laughs) the agent? Like, is it actually saving my time or is it wasting my time? And the, the more often you see it wasting your time and just, um, uh, burning compute cycles, the less likely you're going to use it again. And so if you are predictable, steerable, you know, verifiable and tolerable. If we get to that f- for all four criterias to a certain level, I think we're going to see a, a wide, um, adoption of, of agents.