GitHub is putting AI front and center. Developers are wary

At this point in 2023, It's hard to find a tech company that hasn't embraced generative AI, especially one owned by Microsoft. But what did GitHub CEO Thomas Dohmke mean last month when he proclaimed that GitHub has been "refounded on Copilot," acknowledging the surge of interest in its coding-assistant tool?

Warning bells immediately sounded for a generation of software developers raised on GitHub. GitHub is the home base for an enormous amount of the world's code; it allows individual developers and huge enterprise developer organizations to easily track and manage versions of that code as changes are added and problems arise.

GitHub was originally founded on Git, the second-most famous open-source project Linus Torvalds contributed to modern software development. To those who have been there from the beginning, Dohmke's statement — and the demo of GitHub Copilot Workspace — signaled a changing of the guard, a new era for GitHub that was focused less on improving the core experience and more on ramming a generative AI black box into every nook and cranny of a widely used workflow.

However, in an interview last week at AWS re:Invent in Las Vegas, GitHub COO Kyle Daigle described the idea of "refounding" the company as a nod to how AI will soon impact most aspects of software development one way or another, rather than a massive course change that pushes the goal of improving a familiar and fundamental user experience to the back burner.

"I think it's not that far off to say in the next two years, writing software — the process, that flow — is likely to be completely different," Daigle said. "The refounding idea is not saying open source doesn't matter anymore, (or) developer collaboration doesn't matter anymore. We're signaling that we're in another nexus moment where software development — all-up — is about to change."

AI take the wheel

As the year of the AI boom comes to a close, GitHub Copilot is likely the most widely used enterprise generative AI tool among professional software developers. The tool, which suggests working snippets of code that solve common problems as developers write software, is being piloted or used to write production code at 37,000 organizations with over 1 million paid users, Microsoft said in October.

But GitHub Copilot is used as an extension to an IDE, or integrated development environment, which is more or less a word processor for writing code. Microsoft happens to build the most popular IDE in Visual Studio Code, but while you can integrate Visual Studio Code with GitHub, they tend to be used somewhat separately.

The classic GitHub experience is centered around the review of code; deciding whether or not new code submitted as a pull request is ready for the main codebase, fixing bugs or issues in production code, or assigning tasks to developers. Dohmke and GitHub proposed major AI-fueled changes to that experience with the demo of GitHub Copilot Workspace, and that made a lot of developers raise an eyebrow.

Using text input to get started and a company's internal code base as training data, Copilot Workspace will automatically propose a plan as well as the actual code changes needed to address bugs, issues, or feature requests. "Copilot Workspace is like a pair programming session with a partner that knows about every inch of the project, and can follow your lead to make repository-wide changes from the issue to the pull request with the power of AI," GitHub said in the blog post announcing the new service, which it plans to launch next year.

But while developers have flocked to coding assistants, where they can review the AI-suggested code right as part of the writing process, there was a fair amount of grumbling that GitHub was about to fix something that wasn't broken in favor of improving the core experience. They also worried about being unable to understand how and why the AI assistant came up with those proposed code changes, losing control over the direction of one of their most important assets to a black box; Workspace users will be able to edit those proposed changes, but it's not hard to see time-pressed developers hitting the button and moving on.

Daigle disagreed with that interpretation, but clearly heard the feedback.

"I think that as a developer … we're the first group to disrupt everyone. And then when disruption comes for us, we suddenly find reasons why it can't always work," he said.

Extending the embrace

GitHub has been at the center of the software development workflow for over a decade without serious challengers.

GitLab is growing steadily, but it wants customers to use its homegrown tools for CI/CD and GitHub is more agnostic. Linear has generated a lot of buzz, but is focused more on issue tracking and project management compared to GitHub's core focus on storing and reviewing code.

We're shipping to learn, we don't want to break things and move fast.

Making sure GitHub stays in that position is a key priority for Microsoft, which enjoys a substantial amount of developer mindshare thanks to GitHub and VSCode. Before it was acquired by Microsoft for $7.5 billion in 2018, the company had a poor reputation for shipping new features that developers wanted and needed, and needed a bit of a jolt from the software giant to get moving.

"Microsoft isn't shy about the fact that over time, they want to benefit from that (investment), too, they want to be able to have GitHub be the front door for developers," Daigle said. But GitHub retains the freedom to make its own decisions about how it incorporates new technology — including Microsoft-owned AI technologies — into its products, he said, and the company spent a fair amount of money on a large booth on the expo floor of Microsoft's largest cloud infrastructure competitor last week.

Nobody likes change, but the need to keep existing users happy at the expense of evolving their products and services along with emerging technologies has doomed many enterprise tech companies to a slow demise. It's a tricky balance, and one that Daigle and GitHub will have to navigate as a new generation of developers and startups grow tired of the old ways.

"What we're trying to find is how can you add these AI powered workflows via Copilot into the rest of the GitHub experience where it actually adds value, not just slap an AI button on it or chat experience on it, and it's going to be totally fixed," Daigle said. "We're shipping to learn, we don't want to break things and move fast."