Build what's next on GitHub, the place for anyone from anywhere to build anything.
Join us October 28-29 in San Francisco or online for GitHub Universe, our flagship developer event uniting people, agents, and the world's code.
From starting at GitHub over 10 years ago as a developer to becoming the Chief Operating Officer, I’ve learned that developers are often bellwethers for change across the rest of their organizations. That makes investing in—and learning from—them critical.
As early adopters of new technologies and practices, developers are often bellwethers of business landscape change. That’s why, at GitHub, we believe that the more businesses can understand what developers need to thrive, the better they can support the rest of their organizations.
That’s never been more apparent than with their adoption of AI. While the world is still grappling with its use cases, we’ve found that developers are already using it at scale. In fact, our recent survey of developers found that 92% are already using AI tools at work or in their personal time. By studying their adoption and implementation methods, we can better apply such tools in the workplace for all teams to benefit.
This is why we conducted this survey, detailed below: to get a sense of what developers think about AI technology and how they want to use it. My take is that they want to collaborate to solve real problems and drive impact while they’re crunched for time. The upside? They believe AI will help.
Below, I’ll talk about how to make the most of AI and invest in a great developer experience (DevEx)—the systems, technology, processes, and culture—to help developers drive impact, and improve productivity and collaboration.
Developers rank designing solutions to novel problems within the top four tasks that make a positive impact on their workday. The rest you can see in the charts below: developers want opportunities to engage with end users and drive impact, yet they often spend more of their time waiting for builds, tests, and deployments.
Developers also want to collaborate. In our latest survey, they report working with an average of 21 other engineers on a typical project—and 52% said they work with other teams daily or weekly.
Developers think collaboration is so critical to their jobs that when asked what performance metrics they should be measured by, they ranked collaboration and communication as the top preferred metric—over code quality.
Here’s the thing: developers think AI can help improve collaboration, individual productivity, and solution design. Among the top benefits:
Early research also shows signs that AI coding tools, such as GitHub Copilot, do actually improve collaboration, too. In one study, we found that developers completed code reviews 15% faster when using GitHub Copilot. And Duolingo also saw a 67% median increase in code review times after adopting GitHub Copilot—which suggests developers who work with AI do spend more time collaborating.
If implementing AI means increased productivity, collaboration, and innovation on engineering teams, then there’s a high incentive for organizations to operationalize AI across other teams, too. And understanding how best to do that starts with focusing on a group that’s already using these tools: your engineers.
Developers are now among the first teams at their companies looking to use AI at scale and organizations should take note of how their developers are building with and around generative AI and expand those practices beyond engineering.
At GitHub, we have taken lessons from the developer community to improve the operations of our entire organization, including non-development teams. We’ve also learned a few things from shipping developer-first products, including GitHub Copilot, the world’s first at-scale and most widely adopted generative AI coding tool. Here are three tips for other organizational leaders who want to operationalize AI to lift all teams.
GitHub builds new products in small and agile tiger teams and integrates tight feedback loops throughout the product development journey. We also try to avoid the sunk cost fallacy in which teams continue down a path they know isn’t right because they’ve already invested money and time into it. This agility allows our teams to build efficiently and make decisions without bottlenecks.
In fact, this approach is what led us to develop GitHub Copilot, the industry’s first at-scale AI coding tool. Several years ago, we got access to an AI model from OpenAI that we knew was powerful and something we wanted to put in the hands of developers.
Yet it wasn’t totally clear how to turn this AI model into a product, and we needed a small team of highly collaborative subject matter experts to figure out the best way to do it. Working with a small team of engineers, designers, and researchers made it easy to agree on a focused problem—assisting developers with coding functions in the IDE (integrated development environment)—quicker experiments, and ultimately, a faster time to market. Once you have a minimum viable product (MVP), you can start bringing more people into a given project to continue iterating.
Our survey findings show that developers collaborate with a lot of different people. So, you might be wondering why I’m emphasizing the efficacy of small teams.
Developers rely on cross-functional collaboration to make sure the products they’re building are solving real problems. When collaboration is done right, teams across the organization are focused on the right problems, which leads to better solutions and stronger business impact. The bottom line is that although the software might be written by a small team, developers aren’t building products in silos.
Quality assurance in the age of AI will depend on agile teams
Outputs from large language models (LLMs) are probabilistic, so they don’t always produce the same, predictable outcomes. That means experimentation with them is statistically based.
Because AI is being adopted quickly and their quality control requires a lot of experimentation, companies that want to operationalize AI at scale will need to prioritize efficient workflows comprising cross-functional teams, quick iterations, and tight feedback loops.
My time as a developer taught me that software development is a team sport—and that makes collaboration critical. The high volume of human collaboration from our survey also suggests that an organization’s operational problems are inherently people problems. That means it’s the culture, in addition to the right tools, that’s going to influence the quality of collaboration.
Organizations can learn best practices for creating a collaborative culture from the open source community, and then adapt them to their internal teams in a process known as innersource.
Developers already see the value in innersource. In our survey, almost 90% of developers said that innersource practices improve team performance.
At a high level, innersource makes work more visible and discoverable throughout an organization with practices like publicly documenting workstreams, decisions (including the context around those decisions), and solutions, which for developers often boils down to code.
But these practices aren’t just useful for developers. In our experience, they are very effective in sharing the context around important decisions with the entire organization, encouraging everyone to pitch in and solve a problem, and helping teams avoid duplicative work. In fact, the Pragmatic Engineer has previously pointed out that the best developer companies focus on providing people with context and information to solve key business problems. This contrasts more traditional, hierarchical companies that assign work to be completed.
Innersource also holds the promise of helping teams be more productive and collaborative while positioning them to reach greater benefits from AI. At GitHub, we frequently say if something isn’t written down and easily discoverable, it never happened. We do this to make sure no one is ever limited by the amount of information or context they currently have—including our AI models.
We don’t just prioritize documentation for our developers, either. For example, we recently released an early alpha of GitHub Copilot in GitHub Support, where customers can get their questions answered without waiting for our support team. The release is part of an ongoing effort to bring AI to all teams inside GitHub so team members can focus on the most significant areas of impact.
If you want to position your teams for success today, and for the benefits of AI tomorrow, here are a few innersource practices you can start promoting:
More documentation can help build a better, more personalized AI experience
Looking ahead, adopting innersource practices will continue to advance the capabilities of AI coding tools. For instance, GitHub is currently experimenting with ways to improve GitHub Copilot’s semantic understanding through vector databases, which could create a customized coding experience for developers working in private repositories or with proprietary code. Internal teams that adopt innersource best practices—like producing high-quality documentation and engaging in collaborative decision making—will only improve customized coding experiences once they’re ready to deploy.
Upskilling is important to developers: it helps them stay ahead of technology changes and design more innovative solutions. Upskilling is equally important for other teams—especially as AI makes its way further into the workforce.
So, how do developers approach learning new skills and what can we learn from them?
In our survey, developers consistently rank learning new skills as the number one contributor to a positive workday. Yet, 30% also say learning and development can have a negative impact on their overall workday which suggests that developers sometimes view learning and development as additional work.
It’s not just developers who feel this way. Employees in general find learning and development often calls for unrealistic time commitments, and research shows that upskilling is most impactful when it’s built into a workflow.
How do you create the time and culture for knowledge sharing and learning across the organization? Here’s how we’ve been doing it at GitHub.
We experiment with AI coding tools that can help to directly educate developers within their workflows. For instance, in the technical preview for GitHub Copilot for CLI, there’s an explanation feature that appears after GitHub Copilot suggests a command. The feature explains the function behind the suggested shell command so developers can verify the command against the original problem they’re trying to solve. Likewise, the technical preview of GitHub Copilot Chat enables developers to ask an AI assistant to explain what a code block is supposed to do, debug code, and explain how to do something in a new language or framework, among other things.
AI-powered learning isn’t just for developers either. As more AI tools make their way into the workplace, more teams will be able to ask questions during their workflow and get answers via an AI assistant. And as AI models improve, they’ll be able to offer more personalized answers, recommendations, and outputs based on internal knowledge bases—which drives home the importance of documentation and innersource practices. An active innersource culture can lead to AI-driven tools that integrate consistent learning throughout all workflows.
AI promises significant potential in helping individuals and teams upskill—but computers alone aren’t enough to help someone develop their skills holistically. We encourage senior engineers to spend more time mentoring junior or mid-career engineers in the work that they’re doing instead of treating it as a separate task. In practice, that means if a developer isn’t sure how something works, they’re encouraged to open a pull request for review well before their code is ready versus keeping their uncertainty a secret. This is true for other roles, as well. Our innersource culture encourages people to speak up and create moments for learning. As a result, mentors and leaders across the company teach at moments of need and empower all employees to take ownership of their work. These are key factors to creating a trusting, collaborative environment that generates innovative solutions.
I believe learning and teaching how to do something well—whether it’s writing clean and useful code or communicating clearly with a diverse group of people—will continue to be a human endeavor that requires time. However, AI tools, and the collaborative culture that results from successful innersource and AI adoption, can help people carve out more time for upskilling—and that’s a huge win.
AI is going to change the game for building company culture and trust—from improving collaboration and performance, to enabling upskilling on the job. As companies prepare for AI adoption, organizations of all sizes will benefit by investing in:
Essentially, investing in DevEx—the systems, technology, processes, and culture that all work together to drive developer productivity and satisfaction—will prepare all enterprise teams for AI-powered growth and innovation.
