AI Is Joining Software Teams — Researcher Says the Rules Need Rewriting

A new academic paper argues that software teams are organized the wrong way for the AI era. Today, most companies decide once who does what: who writes code, who checks it, who makes the final call. The author, Javier Cámara, says that setup goes stale fast, because the work, the risks, and the tools keep changing underneath it. His fix: treat the whole team — people and AI together — as something that constantly rearranges itself.

The trigger for this is "agentic AI" — AI that doesn't just suggest things but can actually take actions, like writing code, running tests, or fixing bugs on its own. These tools are spreading quickly through software jobs. But most companies still hand AI a fixed role, like a junior helper who never gets promoted or reassigned. The paper compares this to a sports coach who sets the lineup at the start of the season and never changes it, no matter who's injured or playing well.

His proposal involves two kinds of change happening at once. First, the software itself evolves, as always. Second, the way the team builds it also evolves — who's involved, who holds authority, what information they see, and how quality gets checked. When goals shift or uncertainty rises, the team reshuffles responsibilities rather than pushing ahead with an outdated plan. The author built a small proof-of-concept demo to show the idea can work.

The honest catch: this is early research, not a product you can buy or a method your company can copy tomorrow. The paper mostly lays out challenges and future plans. Still, it points at a real shift. As AI takes on more coding work, the interesting question stops being "can AI write code?" and becomes "who decides what, when an AI is doing the work?" That's a management question, and it will land on ordinary workers long before it lands on engineers.