“19% slowing observed in experienced developers is not a hint of AI as a whole, but a reflection of the actual friction of the integration of probabilistic designs into deterministic workflows,” Gogia explained, stressing that the measures should include “Dowstream, code and cycles for mutual controls and only time to be.”
Wider industrial evidence
The metro findings coincide with the trends identified in the research and evaluation report by Google 2024 Devops (Dora) based on 39,000 experts. While 75% of developers said they feel more productive with AI tools, the data tells a different story: every 25% increase in AI acceptance showed a 1.5% decrease in delivery speed and a system of system stability 7.2%. In addition, 39% of respondents said they had little or confidence in the General-Generald code.
These results are contradictory optimistic studies. Research MIT, Princeton and University of Pennsylvania, analyzing data from more than 4,800 developers in Microsoft, Accenture and other Fortune 100, found that developers using Github Copilot completed an average of 26% more tasks on average. A separate controlled experiment found that developers have completed the coding tasks by 55.8% faster with Github Copilot. However, these studies usually used simpler, more isolated tasks compared to complex real -world scenarios that are examined in a subway research.