AI coding agents are making dev teams more siloed, not less
The pitch for AI coding agents is that they act like new teammates: taking on tickets, writing code, and freeing engineers to do more collaborative work. A study of real-world usage suggests that is not what is happening.
What the data shows
Researchers analyzed 25,264 agent-generated pull requests across 2,361 popular GitHub repositories over a three-month stretch in 2025. The numbers are not flattering for the teamwork narrative.
- The median project produced just one or two agentic PRs over the entire period.
- In 70% of projects, fewer than one in five contributors participated in an agentic workflow.
- In 79% of agentic PRs, the same developer both reviewed and modified the AI’s output. No second set of eyes.
- Only one in eight agentic workflows involved more than one human.
Why it plays out this way
Courtney Miller, assistant professor of computer science at George Washington University, points to how agents are being adopted. Each engineer develops their own prompting and correction habits, which means teams that start from the same place quickly drift into very different processes.
Software engineering consultant Sarah Wells has seen developers turn to agents for conversations they would once have had with colleagues. For junior engineers, that is a specific risk: they may not have enough experience to recognize when an agent’s suggestion is wrong.
What engineering leaders should do
Miller argues that leaving each developer to figure out agent productivity on their own is the wrong model. Teams need shared learning spaces: Q&A channels, brief stand-up discussions, and live demos using their actual systems. Wells has seen organizations use AI-focused hack weeks and showcases to spread knowledge, but notes that engineering culture determines whether people share failed experiments as freely as successful ones.
Leaders also have to measure quality, not just volume. Code count and PR totals are easy to game and say nothing about whether the output is secure or maintainable. Testing and quality assurance need to scale alongside code generation, or the agent just moves the bottleneck into the review queue.
The fix, in short, is treating agent adoption as a team-level capability problem rather than an individual one.
I’m founder of BizStack at Cagri Sarigoz LLC and a passionate advocate for entrepreneurs.
With over 14 years in tech, marketing, and AI, including my role as Head of SEO at CitizenShipper and co-founder of TaleBot at Intale AI, I’m dedicated to sharing genuine, useful product insights and tips.
At BizStack, I aim to cut through the digital noise to provide clear, actionable advice.
And more than all else, I’m a father to a (always) little girl and a husband.
Contact me at [email protected] for assistance.
