Generative AI tools like GitHub Copilot are not just speeding up coding; they are quietly rewiring how software developers divide their time between writing code, managing projects, and learning new skills.If you think about it for a moment, it only makes sense that as developers start using AI more, it would change how they spend their time at work. The question is: “How?” Now, thanks to a Harvard Business School working paper, “Generative AI and the Nature of Work,” which surveyed 187,000 open-source developers, we’re getting answers.The researchers found that programmers given free access to GitHub Copilot significantly shifted their work mix toward hands‑on coding. Before Copilot, developers spent about 44% of their time on coding tasks and 37% on project management, such as triaging issues, reviewing pull requests, and handling support queues. After Copilot arrived, coding time rose by 12.4%, while project management activities fell by 24.9%.Not all such studies agree with this result. For example, Google’s 2025 DORA State of AI‑Assisted Software Development found that while people using AI produced more code, more quickly, the code also came with more problems, which required more bug-finding and rework.Other studies agree with this. For example, the development tool company Sonar‘s 2026 State of Code Developer Report, which surveyed more than 1,000 developers, states that 96% of developers have trouble trusting AI‑generated code. Further, 38% say reviewing AI‑generated code requires more effort than reviewing code written by colleagues. In the same report, “correcting or rewriting code created by AI coding tools” is listed as a top source of toil, especially among frequent AI users.One of the Harvard report’s co-authors, Frank Nagle, Research Scientist at the Initiative on the Digital Economy at MIT, tells The New Stack that in “the current study, we did not drill into the quality of the code. But, anecdotally, I’ve heard similar things.” So, when people say they’re “coding” today, it’s not necessarily what we’d call coding only a few years ago.Crucially, the researchers interpret this to mean that AI is not just a productivity bump, but a change in developers’ jobs. For maintainers and top contributors who keep high‑profile projects running, this rebalancing matters greatly. The study notes that these “power users” often spend a disproportionate share of their time on secondary tasks such as reviewing others’ contributions, merging code, and resolving user‑reported issues. That’s often a fast track to burnout.Specifically, developers using Copilot saw their project‑management activity increase by nearly a quarter, while peer collaboration events decreased by almost 80%. Why? The authors think it’s because they needed to do less debugging and refactoring. That shift suggests that generative AI is acting as an always‑on reviewer and pair‑programmer, handling many of the small, routine fixes that previously would have required human intervention.That’s good and bad news. While this relieves workload pressure, it also hints at a cultural risk for open source. Nagle warns in the paper of a “retreat away from teamwork” as developers lean on AI instead of colleagues for advice, design feedback, or code review. What has long distinguished open source is not just publicly visible code, but the dense web of human collaboration that shapes both software and developers.While time‑saving is real, the loss of human interaction erodes the social and organizational value that companies and open-source communities derive from collaborative work. It appears that AI, by reducing peer contact, is another force hollowing out the human side of software engineering.Another problem with this drastic decline in people working together is: “What happens when code changes flow with fewer human checkpoints?” Amazon has recently been showing us a real-world example of what happens: Nothing good. Due to numerous AI failures, Amazon is now requiring senior developers to oversee and approve the work of junior and mid-level AI-assisted programmers.At the same time, the Harvard report found that the largest incremental benefits from generative AI appear among less‑experienced workers. Within the GitHub sample, lower-level developers using Copilot saw the largest increase in time spent on core coding activities. That said, as Nagle points out in The New Stack conversation, “Everyone in the sample is a maintainer, so even the lower-experienced people still have some higher level of experience than a total rookie.”That said, in the paper, Nagle explicitly calls it a “profound strategic error” when companies cut junior hiring on the assumption that AI can fill the gap. The paper found that AI works best as a complement, accelerating skill development and preparing workers for higher‑level responsibilities. “When companies stop hiring entry‑level people, it’s short‑term thinking at the expense of investing in the future.”Another result is that the Copilot-enabled developers increased their cumulative exposure to new languages by nearly 22% relative to baseline. This suggests that the tool enables low‑cost experimentation. Thus, AI is making it easier to try out a new language or framework because the AI can fill in idioms, boilerplate, and syntax on the fly.What it all boils down to is that AI is clearly changing the very nature of programmers’ work. It also appears, though, that, just like AI itself, it’s still too early to say exactly what those changes will be. Strap in, folks, it’s going to continue to be a bumpy ride.The post GitHub Copilot’s effect on collaboration has stunned researchers appeared first on The New Stack.