A recent study challenges the prevailing narrative that AI tools universally boost workplace productivity, particularly for skilled professionals. Conducted by researchers from the nonprofit Model Evaluation and Threat Research (METR), the experiment involved 16 experienced software developers performing 246 real-world tasks. While the developers predicted AI assistants like Cursor Pro and Claude would reduce their task completion time by an average of 24%, the opposite occurred: using AI actually increased their time by 19% compared to working without it. The researchers found that developers spent significant time cleaning up, debugging, and retrofitting AI-generated code to fit the specific context and architecture of their existing projects, ultimately slowing their workflow.
The findings highlight a critical gap between AI’s capabilities and the complex, nuanced reality of expert work. Experienced developers bring years of accumulated context and problem-solving strategies that current AI tools lack. As study participant Philipp Burckhardt noted, the technology may have even hampered his efforts. This aligns with broader economic research questioning the scale of near-term AI productivity gains. For instance, University of Chicago professor Anders Humlum found only a modest 3% productivity improvement among AI-using workers in Denmark, while MIT economist Daron Acemoglu estimates that only 4.6% of tasks in the U.S. economy will be made more efficient by AI, warning against the rush to automate processes that shouldn’t be automated.
The study’s authors caution against sweeping conclusions, noting the small sample size and the fact that the tools were new to the participants. They emphasize their goal is to „pump the brakes” on torrid AI implementation, advocating for more high-quality measurements before making high-consequence deployment decisions. The research underscores that successful AI integration requires more than just technology—it demands organizational adjustment, complementary investments, and thoughtful consideration of when to leverage human expertise versus automation. For now, the case of the slowed-down developers serves as a potent reminder that for skilled workers, the promise of AI as a pure productivity booster remains unproven and highly context-dependent.
Ez a cikk a Neural News AI (V1) verziójával készült.
Forrás: https://finance.yahoo.com/news/experienced-software-developers-assumed-ai-154445703.html.