How AI Is Shifting Developer Culture and Work at GitHub

Among all the fear, uncertainty and doubt in the wake of generative AI, GitHub has found something that’s surprising: AI is making its developers happier.

Software developers have too much to do, said Inbal Shani, chief product officer at GitHub. Coding may be the job, but along with it, developers now have to write documentation, create tests, run tests, meet with internal and sometimes external stakeholders, perform code reviews, handle systems architecture, and — oh yes — debug existing code.

“The focus of AI through the lens of a developer is really about productivity,” Shani said. “The developers are making it into a table stakes within the industry and they want to use it because they understand the optionality that it gives them, they understand the flexibility that these tools have for them and the potential advantages that they can get from that.”

The Industrial Revolution of Software Development

AI is ushering in the “Industrial Revolution of software development,” Shani contended because it fundamentally changes how developers write code as well as how developers think about software development.

Time is just one measure of developer productivity, she said. As measured by GitHub, productivity is also about reducing the load developers have and reducing stress.

“When we started testing AI within GitHub, we started seeing developer happiness go up,” she said. “The junior developers, Copilot became that pair programmer that works with them through the beginning while the senior developer is there to help. We see the senior developers getting more time to focus on system architecture and system design, and the more tricky testing that they need to do or spend more time thinking about the documentation versus just writing it.”

That’s probably why 92% of developers report they’re already using AI, she added. In Microsoft’s January earnings report, GitHub shared that 50,000 organizations and 1.5 million developers have deployed Copilot.

“It’s likely the fastest scale adoption that we’ve seen for any developer tool or transformation in the past seventy years of writing code,” she said. “So it must be that this magical unicorn, AI, is doing something right. And the biggest thing that it does, right, is that when I’m looking at Copilot, it just solves the really critical challenges that developers have and it takes away that burden from them so they can focus on the things that matter, which is actually the complexity of writing code and the complexity of thinking about system architecture.”

AI Adaption

AI is still in its infancy, relatively speaking. When Shani came into IT, she was an applied scientist and developed algorithms for specific challenges to solve AI. It was still a niche solution, she said. But 2023 was a transformative year where AI democratized.

“We started bringing more AI capabilities to software development on a scale that has never been seen before,” she said. “What we’ve seen in the past year and a half, or close to two years, is that we took the concept of AI from that magical black box that only specific people can know how to tune, and we made it into something that is more available for all software development.”

AI also will change software culture, she said. Indeed, it already is, she added.

“If you remove some of those stress factors because you’re making the developer much more productive, then you’re creating a happier, more productive environment for that organization to thrive,” she said.

Like the Industrial Revolution changed work, so will AI. For instance, it used to be enough for a junior developer to focus on learning coding and organizational coding conventions during their first few years. But AI can get junior developers onboarded more quickly by teaching them on what they need to know in terms of how the organization thinks about code, syntax and naming conventions, expectations and guardrails, she said.

“We’ve already seen several of our customers that are using Copilot to help onboard new developers, as [in] here’s how you write code in the company,” Shanis said.

Changes GitHub Sees

Generative AI empowers developers to be more creative about what they create with code, because they can program faster and automate tasks they previously had to handle manually.

“You don’t have to use Copilot to write your entire code, you can choose where you use Copilot, and these are usually the areas that you don’t enjoy doing,” she said. “It’s really their own choice, and this is why they’re the pilot and Copilot is their copilot.”

Developers can adapt by applying generative AI to help where and when they need it, or for tasks they currently don’t favor, or where they just haven’t always had the time to do the task well. For instance, one area where GitHub is applying Copilot is to help identify when there’s a security issue with a code set, she said.

“We’re also going to suggest how to fix the vulnerabilities … or which repos to avoid because there is detected vulnerability, so we are seeing the big element of AI that is growing beyond what Copilot was originally,” she said.

Boot camps and university programs also will need to adapt, she added. For instance, developers used to just learn code and focus on that for the first few years. While developers will still need to start with the fundamentals of coding, they’ll also be expected to learn systems architecture and design earlier. Perhaps more importantly, the AI tools will become a regular part of the curriculum because that’s the new standard of writing code, she added.

“We need to teach developers — frontend developers and backend developers — how to use it,” she said. “Yes, they have to understand code, they need to understand code… but it’s also going to move the need for system design and system architecture to much earlier in their career.”