What has changed for software developers in the past 12 months?
A year ago, we were talking about AI helping to write code. A year ago, developers were using AI but it was a very one-to-one relationship. I was going to write some code, so I was going to ask AI to help me. Now, we are in a world where not only am I, as a developer, working across a multitude of GitHub Copilot, CLI, VS Code with Copilot, I’m working on three projects at a single time. It is not just about writing code anymore. It is about operating and deploying it. It is about testing it or letting the agents test it on my computer. It is really expanding the role of what a developer can even do. A single developer can meaningfully go from idea to production and scale by using agents and skills, which we weren't really talking about last year. We were just talking about how much code AI agents can write for me. Now, we are trying to find the upper limit of what an agent can do in service of my idea as a developer.
What is the upper limit for AI agents in the software development life cycle (SDLC)?
I think developers are pushing the upper limit, when we look at some of the open source projects that have been built or grown significantly in this era and in India, such as Hyperswitch, ERPNext, Tooljet and Bruno. We use flows at GitHub that see a bug and immediately try to fix it or go find more documentation. Automation workflows — Copilot CLI in particular — enable developers to close the loop much, much more quickly rather than focusing only on the code-writing process, which is only about 20 per cent of a developer’s time anyway. The rest of it is all about operating and deploying. Agents are much more capable of going further up that stack and through the GitHub platform.
Experts say that traditional software development is dead with the coming of AI agents. How are agents disrupting the software development cycle?
In the earlier days, we talked about the waterfall model for SDLC. We write all the specifications and then we go to the next step, all the way down to production. I think when we are building software with agents the steps are no longer clearly defined. You don’t finish code and then go to test it. Instead there is an agent testing it in the background as you’re writing it. It is really about how you define the steps — when the steps can now happen concurrently, instead of consecutively. We talk about build, deploy and operate in this new model because it tends to be where our focus is. While I am building, it is running tests in the background and checking if any new requirements are needed. The essence is developers are no longer doing one task and handing it to the next person. Instead, they’re able to move more quickly and get more done across the SDLC. What developers are experiencing is SDLC is now more overlapping and the stages are not as distinct as they once were.
How does the role of a developer change?
What we are seeing at GitHub is more and more technically-minded people are doing the work with these agents in ways that we may assume are only intended for developers. I have folks in my teams who are not developers but are building end-to-end features inside the GitHub monolith because they are able to use the feedback from open source maintainers.
Hence, we are seeing this idea of developers and builders kind of coming together in this era. We, at GitHub, are talking about our goal of a billion developers and I think that agents are making it more accessible for people to explore software development even if they don't want to make it their career.
 "Kyle Daigle, chief operating officer, GitHub")