The surprise findings of a new research found that coding artificial ielligence assistas not only increase the speed of experienced developers, but they actually finish their work by 5 %. This study shows that despite the positive notion of programmers themselves, the functional reality of these tools in complex projects is quite differe.
Although artificial ielligence tools have conquered the technology world’s productivity, the new study of the nonprofit organization Metr The results were completely unexpected. The study, which focused on professional developers, showed that the use of smart coding assistas reduced their speed.
Reduce developers’ speed due to the use of artificial ielligence
In this study, four experienced open source programmers performed two actual task (from bug fixes to new feature developme) in large projects they were fully familiar with. They were randomly allowed to be of artificial ielligence tools (mainly Cursor Pro With models Claude 3.5 Or 3.7 SONNET) To use or not.
The results of this experime revealed an ieresting coradiction; Prior to the start, developers predicted that they would be 5 % faster with artificial ielligence. Even after the study, they also believed that they had 5 % improved productivity. But recorded data showed that developers when using artificial ielligence on average 5 % more Have spe time completing their duties.
But Why does artificial ielligence slow down? Researchers have ideified several reasons for this:
- Excessive optimism: The curre capabilities of artificial ielligence have not yet advanced as well as users’ optimism.
- Familiar with the code: Experienced developers were so domina over their codes that artificial ielligence could not offer them a significa shortcut.
- Project complexity: Artificial ielligence in massive projects (with over one million lines of code) has weaker performance than small and limited issues.
- Uncertaiy to offers: Programmers accepted only less than 5 % of the codes produced by artificial ielligence and spe a lot of time reviewing, validating and modifying its suggestions.
- Lack of understanding of the field: Artificial ielligence had difficulty understanding the coext and hidden relationships in large projects, and often offered irreleva suggestions.
Despite these results Should artificial ielligence be abandoned? The answer is “no”. Researchers insist that these findings should not be generalized; These tools may be very useful for less experienced developers or in smaller projects.
Ierestingly, many participas in this study coinue to use these tools despite seeing the results. Their reason is that, although not always faster, artificial ielligence can reduce the meal stress caused by some repetitive and boring aspects of coding and make this process less frustrating. This study shows that at prese, smart assistas play more of the role of a “partner” that requires careful monitoring, and is not a tool to increase flawless speed.
