Opinionated Read: How AI Impacts Skill Formation (www.sicpers.info)

A new preprint from authors associated with Anthropic presents findings that suggest AI usage can hinder essential coding skills like conceptual understanding, code reading, and debugging while not significantly enhancing efficiency. Although some participants experienced productivity gains by fully delegating coding tasks to AI, this came at the cost of skill acquisition, particularly concerning learning a specific library. The study identifies six distinct patterns of interaction with AI, noting that three of these patterns maintain cognitive engagement and resulting learning, urging caution when integrating AI tools in professional workflows—especially in safety-critical domains like software engineering. The research highlights crucial technical implications; using AI for tasks didn’t yield meaningful improvements in completion times, and those who utilized AI performed worse on subsequent learning assessments. Notably, developers with limited Python experience benefitted from AI assistance without sacrificing their understanding of the library, raising concerns about participants' overall coding competency and retention of knowledge gained. Furthermore, the authors found that one AI interaction pattern—iterative debugging—proved less effective, underscoring the limitations of language models in evaluating code logic. Overall, while AI tools can enhance productivity, their role in skill development remains contentious, necessitating further investigation into their long-term impact on professional competencies in software development.