Photo: The thing made to help you is now a hindrance to your performance. Ironic
There is a new study that destroys the idea that artificial intelligence super charges the work of experienced software developers. Instead of sharpening them, experienced coders were really slowed down when they were dealt with projects when they were already well aware of the projects.
This study is done by AI Research Non -Popular Met R and is focused on experienced developers working with a famous AI coding assistant cursor, which he was aware of. Before diving into the study, the developers expected that AI would save their time, guessing that it could reduce the completion of work by about a quarter. Even after the use of AI, many people still felt that it had accelerated them about 20 %. But strict data told a different story: AI extended the time needed to eliminate the tasks by 19 %.
Joel Baker and Net Rush, who guided the research, admitted that the result was caught as a result. Rush, before the study began, had predicted that AI would double the production capacity – and that would make many of us think. However, these results have raised suspicion of the widespread assumptions that AI tools reliably promote the production capacity of highly skilled, high -cost software engineers. This idea is something in which many companies have chosen to invest a lot and have recently been invested. This comes when some tech leaders, including Dario Amody, CEO of the AI company, Entropic, have suggested that AI can eliminate half of all half -admission level collar jobs over the next five years. Previous studies have increased the hype, which reports the key benefits of productivity: One claims that AI has increased coding by 56 %, while the other has completed 26 % more tasks within the same time.
But a meter study tells a very different story. When the developer is working on large, complex codes bases, the productivity does not maintain, which he knows well. In these cases, AI not only failed to help but actively slowed the developers. This problem was born by the need to double check and often need to correct AI’s suggestions-such suggestions that were often nearing to be accurate, but are not so precise to trust without careful review.
Baker explained that the video recording of the participants shows how the AI often bowed the developers in the right direction, but rarely needed to supply it.
When we saw the videos, we found out that AIS made some suggestions about his work, and these suggestions are often correct in the direction, but not at all.
– Joel Baker, Meter Research Leader, July 2025
This resulted in extra -time review, amending and sometimes rejecting AI’s output.
Researchers were cautious that these results did not apply to the entire board. Low experienced engineers or workers at an unfamiliar code base can still benefit from AI aid.
What about smartphone AI?
Photo by Apple
The results of this study can also show a glimpse of how everyday smartphone users (such as you and I) interact with AI -powered features on their devices. Like developers, many smartphone users expect AI tools-from articles and voice assistants to photo editing tips-to smooth daily tasks.
But in these features we often need to stop, review and correct AI’s memories, sometimes simple steps make them feel much more complicated. Whether it’s an automatic error, an increase in a faulty framed image, or the AI-Infield message response, we are also discovering as a smartphone users that the AI is far from flawless.
The convenience can sometimes come at a cost!
Hold on to the Surfers VPN now at a distance of more than 50 % and with 3 additional months!
Save your connection now at the expense of deals!
We can earn commission if you make purchases
Check the offer
Read the latest from Sebastian Pierre
