Software Engineer Workflows have changed in recent years by the surge of AI coding items Converge And GitHub Copilot, which promised to develop productivity by automatically writing lines of code, repairing bugs, and test changes. Tools are run by AI models from Openi, Google Defermind, anthropic, and Xai with rapidly increases their performance to a range of software engineering tests in recent years.
However, a New study Published Thursday in non-proof research meters calling how AI coding items are coming today for productivity for experienced developers.
Metr conducts a randomized controlled test for this study by recruiting 16 experienced open source and have real-work functions. Researchers randomly handed in half of the tasks as “AI-real,” giving the author’s permission to the cursor functions, while other cups of AI use.
Before their assigned assignments are finished, developers predict the use of AI tools to AI reduce the time to complete 24%. That’s not the case.
“Surprisingly, we know that AI’s permission of the actual increase in 19% – slow developers when using AI tool,” as researchers.
Especially, only 56% of study developers have experienced using the cursor, the main AI tool offered to study. While almost all developers (94%) have experience using some web-based llms in their coding workflows, this study is the first time using some used cursor. Researchers notice that developers are trained to use the cursor in preparation for study.
However, MECR’s findings apply questions about the Productive Products AI-promised in 2025. Based on the Deviling AI – Based on what is known to “vibe coders”
Metr researchers point to a few potential reasons why ai slowed down developers rather than speeding them up: developers spend much more time prompting ai and waiting for it to respond when using vibe coders rather than actually coders. AI also wants to struggle with large, complex code bases, using this test.
The authors of the study are careful not to get any strong conclusions from these findings, clearly they do not believe AI systems to fail to accelerate many software developers. other Many studies It is shown that AI coding supplies facilitate software engineer workflows.
The authors also noticed that AI developed many years and they could not expect the same results for three months from today. MetR also knows that AI coding items that have improved their ability to Complete complex, long-term assignments in recent years.
However, researchers offer another reason to have no doubt of the promised profits of AI coding items. Other studies show that AI equipped with AI coding introducing errors and, in some cases, Security tights.
