AI pair coding: Fast, but devs don't question the bot

Researchers at Saarland University found that developers using AI coding assistants like GitHub Copilot learn as much as those in traditional pair programming sessions, but show concerning levels of trust in AI suggestions. Developers frequently accept AI-generated code with minimal scrutiny, potentially reducing opportunities for deeper learning from broader discussions between human partners. While AI pair programming increased efficiency for simple tasks, it limited spontaneous knowledge exchange typical in human collaborations, raising concerns about long-term learning outcomes.