A study published by Anthropic examined how software engineers acquire new knowledge about libraries when working with AI tools versus without them. The AI-assisted group completed their tasks more quickly, but achieved only 50% mastery on follow-up quizzes measuring concept understanding — while the manual group reached 67% mastery. The findings raise important questions about how companies design AI-assisted learning environments, suggesting that using AI for debugging and explanation while keeping logic and architectural decisions as human tasks may lead to deeper, more lasting skill development

As AI becomes embedded in daily workflows, organizations need to think carefully about the difference between productivity and proficiency — because speed without understanding can create long-term capability gaps.