Despite the hype around artificial intelligence as a productivity booster, a new study reveals that AI tools may actually slow down experienced software developers—especially when working on codebases they know well.
The study, conducted by AI research nonprofit METR, examined how seasoned developers performed when using Cursor, a popular AI coding assistant, on open-source projects they were already familiar with. Surprisingly, instead of speeding things up, AI increased the average task completion time by 19%.
Before the study began, developers anticipated a productivity boost, estimating the AI would cut their task time by 24%. Even after using the tool, they believed it had made them 20% faster. In reality, it had the opposite effect.
“We were shocked,” said co-author Nate Rush, who admitted he initially expected “a 2x speed up, somewhat obviously.” The findings cast doubt on the widespread assumption that AI tools universally enhance developer efficiency—a belief that has driven significant investment in AI-powered software development tools.
The results also add nuance to the broader conversation around AI and job disruption. While some executives, like Anthropic CEO Dario Amodei, have warned that AI could eliminate half of all entry-level white-collar jobs within five years, the METR study suggests that experienced professionals may not benefit as much—or as immediately—as expected.
Previous studies have painted a more optimistic picture: one reported a 56% speed increase with AI, while another found developers completed 26% more tasks. But the METR study suggests those numbers don’t tell the whole story. Many prior experiments use simplified benchmarks that don’t reflect the complexities of real-world software development.
The slowdown, according to METR, came largely from the need to review and correct AI-generated suggestions. “The AI’s suggestions were often close to correct, but not quite right,” said co-author Joel Becker. “Developers had to spend time double-checking and fixing the output.”
Importantly, the researchers note that these results likely don’t apply across the board. Junior developers or engineers working in unfamiliar codebases might still see significant productivity benefits from AI assistance.
Interestingly, despite the slowdown, most participants—and the study’s own authors—continue to use Cursor. The reason? It makes coding feel easier and more enjoyable.
“Developers have goals beyond just finishing as fast as possible,” Becker explained. “Sometimes, using AI feels like editing a draft instead of starting from a blank page. It’s less effortful and more pleasant.”
We have helped 20+ companies in industries like Finance, Transportation, Health, Tourism, Events, Education, Sports.
