AI coding tools may not speed up every developer, study shows

Software Engineer Workflos has been converted by AI Coding Tools in recent years by arrival Cursor And Githab Kapilot, which automatically write the code line, promises to increase productivity by fixing bugs and writing test changes. Tools are driven by OpenAI, Google Dipmind, Ethnographic and Jai’s AI models that are included Quickly increase their performance In recent years of various software engineering exams.

However, a To study new Today’s AI coding equipment published by non -profit AI Research Group Motors calls on how much question that increases productivity for experienced developers.

METR 16 hiring an experienced source developer and conducts a random controlled test for this study by completing 246 practical work in large codes storage that contributes them regularly. Researchers randomly presented about half of these tasks as “AI-Manjur”, allowing developers to use sophisticated AI coding equipment like Cursor Pro, while on the other hand, the other half work prohibits the use of AI equipment.

Before completing their scheduled tasks, the developers predicted that using AI coding equipment would reduce their completion time by 24%. It wasn’t.

“Surprisingly, we can see that AI is actually extended by 19%to finish – developers are slow when using AI tooling,” researchers say.

Significantly, only 56% of the research was experienced using the proposed original AI equipment cursor in the study. Although almost all developers (5%) have experienced some web-based LLMs in their coding workflow, this study was for the first time some used cursor. Researchers have noted that developers were trained to use cursor to prepare for study.

Nevertheless, METR’s search raises questions about gaining universal productivity committed in 2025 by AI coding equipment. Based on the study, developers should not assume that AI coding equipment – especially known as “Vib Coder” – will speed up their work flow immediately.

METR researchers have indicated on several possible factors why AI developers have slowed down the speed.

First, the developers spend a lot more time to request AI and actually wait to respond when using a sibling coder instead of coding. AI also tends to fight in large, complex code bases, which used this test.

The authors of the study are warned not to make any strong decisions from these inquiries, clearly noticed that AI systems have failed to speed up many or maximum software developers now. Another Study to large sizes Show that AI coding equipment speeds software engineer workflies.

Writers have also noticed that the progress of AI in recent years was sufficient and they will not expect the same results even three months from now. METR also found that AI coding equipment has improved significantly in their skills Completely complex In recent years.

The research provides another reason for being skeptical about the promised profit of AI coding equipment. Other studies have shown that today’s AI coding equipment could The incorrect introduction, And in some cases, VulnerabilityThe