A recent study from UC San Diego and Cornell University, released on January 5, 2026, reveals that experienced software developers remain skeptical of AI agents capable of autonomous code generation. Contrary to the “vibe coding” approach, these developers emphasize maintaining control over AI tools through strategic planning and validation. Pedro Domingos of the University of Washington ignited debate by stating that AI coding tools fail in handling business logic or existing code. Surveying 99 seasoned developers, the research found that while AI agents can enhance productivity for simple tasks, complexity and integration with legacy systems severely limit their utility. Quality attributes such as correctness and maintainability are prioritized, with many developers modifying agent-generated code frequently. Although AI agents serve as valuable collaborators, they cannot replace human expertise; oversight remains critical to ensure robustness and reliability in software development. This study underscores the essential role of developers in leveraging AI while guarding against over-reliance.
Source link
Share
Read more