GitHub wants AI to help train the next generation of coders

Of all the jobs that generative AI could change in the future, one of those most impacted might hit close to home for the tech’s creators: software engineering.

Code repository GitHub is leading this change with its Copilot AI programming companion, which seeks to assist coders in all sorts of ways, from suggesting autocomplete-style snippets of code to helping write documentation. And as GitHub’s VP of communities, Stormy Peters is tasked with helping developers understand and use tools like these.

We caught up with Peters to talk about how AI is changing coding, how it can play a role in educating the next generation of software developers, and the challenges around funding open-source programs.

This interview has been edited for length and clarity.

What does AI mean for communities and developers on GitHub?

So AI is obviously a hot topic and developers in general really like it—they’re really excited about it. It’s new technology, and new technology is fun, but it also helps them do their job better. We know that developers that use Copilot are happier, and they write code 55% faster. They don’t have to remember all the syntax or write out all the boring code; they can just get Copilot to do those parts for them, and they can focus on how the pieces go together. So it’s really a “wow” factor.

Where I think it’s really interesting is the people learning to write code in the education space—how do we use this AI to help them learn faster? We’ve been working with professors and educators to understand what this means for them. And it’s been a really interesting journey. Because in the beginning, they were like, “Oh, it’s helping them cheat; turn it off. How do I turn this thing off? How do I get rid of it?” They felt a little bit like that with code that was open-source on GitHub too, like, “Can you make this so my students can’t see that?”

But we’ve really seen them turn around just in a few months. So professor David Malan at Harvard works with us a lot. And he [initially] turned off Copilot for his students, and now he’s writing his own version of Copilot to help the students learn how to code. We’re finding it’s really good for new users. It’s not their main teacher, but a secondary teacher, a tutor. They can ask the same questions a bunch of times. They can say, “How would you write this? How would I do this?” And they still have to know how it fits together, but it helps them along that journey.

How much is AI going to change the future of coding?

It’s gonna let us tackle much harder, more advanced problems. I feel like open-source did that. So before open-source was around, you had to develop pretty much everything from scratch every time you wanted to write anything. And now we have these libraries and massive sets of code that already do so much of the work for you that you can focus on the problem you’re trying to solve and build on top of all those other solutions. And I feel like AI is going to do the same thing. So one of the Duke professors—professor Noah Gift—said he’s seen the students go from, like, he’s not trying to teach them a language anymore. Like, they can almost use any language with the help of an AI, and he has them focused on projects instead of just algorithms or just writing snippets of code.

How does that change the skill set that you need to be a developer?

I think it raises the barrier. I think all good developers have the skill set. But now we’re getting there faster, understanding how pieces go together and understanding the bigger problem you’re trying to solve for the user. And we’re getting there much faster because you’re spending less time just learning. I remember when I first started writing Fortran, you’d have to have so many spaces in the beginning of every line. And that killed me for, like, a half an hour to figure that out.

How are you looking at the shorter-term future? What’s the next step?

We’re working with professors right now to try to understand how they want to integrate it in a classroom and what that looks like and what tools we can offer them…So trying to figure out how they’re using it and how we can make it even better for them. Both developers and educators and learners. On the student side, we added experiences where, instead of just saying, like, “Here’s GitHub, go create a new repo and write something,” we asked them, “What are you trying to do? Are you trying to create a website or are you trying to do one of these following projects?” And then we have a recipe book they can walk through. I think one of the biggest problems is people need a problem to solve in order to learn how to code.

Outside of the world of AI, what else is GitHub working on?

We’re trying to find more ways to fund open-source software developers. When we got started 20 years ago, we thought companies would hire all those people. And now we’re seeing these pieces of software that are used by…hundreds or thousands of companies, and three people write the code. So obviously, like, all those companies aren’t going to hire somebody. So all those companies could donate $10 a month or something to these developers that maintain it. So we’re looking at new ways of funding, sort of like sponsors that you can give money to developers. We have an accelerator, which is a way of helping open-source software developers find their next career path. We’re working with Microsoft’s venture-capital fund to provide venture-capital money to open-source software projects. As it turns out, venture capitalists love to see code and open-source as a proof of concept. But they don’t love to fund something that you’re giving away.

What are some of the challenges for the open-source model right now?

I think the challenge is funding. So that’s why we’re working on the sponsors, the accelerator, the venture-capital plans. The other for them is getting new maintainers. So hopefully, we can help connect these people who want to learn with projects that are willing to mentor them. Because it does take some time to mentor a new user; you have to be willing to give it upfront to get more help later.