AI might be weakening core coding skills, Anthropic study finds

A new internal study at AI company Anthropic has found that engineers using its Claude AI are concerned about the decline of their fundamental programming skills, even as their productivity increases.

The research, based on a survey of 132 engineers and 53 interviews, revealed a significant tension. While employees reported using Claude for nearly 60% of their work with a 50% productivity gain, a notable concern emerged about skill retention. Engineers specifically worry that over-reliance on AI assistants leads to the weakening of the deep, hands-on coding abilities developed through manual problem-solving.

This decline in skills is considered problematic because effectively using AI requires strong supervision and validation of its output, a task that depends on the very expertise that may be diminishing. In the study, some engineers reported deliberately practising tasks without AI to maintain their proficiency.

The study also found that Claude enables engineers to complete 27% more work, often in areas outside their primary expertise, making them more “full-stack.” However, this broadening of capabilities exists alongside the anxiety about losing foundational coding skills. The research from the AI developer suggests these professional challenges may foreshadow similar shifts across the tech industry.