> When developers are allowed to use AI tools, they take 19% longer to complete issues—a significant slowdown that goes against developer beliefs and expert forecasts. This gap between perception and reality is striking: developers expected AI to speed them up by 24%, and even after experiencing the slowdown, they still believed AI had sped them up by 20%.
Fascinating that their perceptions basically matched their predictions, and were totally off. Take any claims of "Opus makes me 10x more productive!" with a massive grain of salt, considering they probably read that on X or HN and then their predictions influence their perception.
The pre-paywall stuff makes it sound like this is about the 2025 METR study of time taken versus time reported. [0]
While the authors noted that they cant prove their experiences are broadly representative of the entire craft/industry, IMO that's a pretty high bar for anybody to clear.
> When developers are allowed to use AI tools, they take 19% longer to complete issues—a significant slowdown that goes against developer beliefs and expert forecasts. This gap between perception and reality is striking: developers expected AI to speed them up by 24%, and even after experiencing the slowdown, they still believed AI had sped them up by 20%.
Fascinating that their perceptions basically matched their predictions, and were totally off. Take any claims of "Opus makes me 10x more productive!" with a massive grain of salt, considering they probably read that on X or HN and then their predictions influence their perception.
The pre-paywall stuff makes it sound like this is about the 2025 METR study of time taken versus time reported. [0]
While the authors noted that they cant prove their experiences are broadly representative of the entire craft/industry, IMO that's a pretty high bar for anybody to clear.
[0] https://metr.org/blog/2025-07-10-early-2025-ai-experienced-o...