METR study finds AI coding tools slow down experienced developers by 19%, contradicting common claims of speed boosts.
The nonprofit AI research group ran a randomized controlled trial with 16 seasoned open-source developers. They tasked them with 246 real-world coding jobs from their usual projects. Half the tasks allowed use of AI coding tools like Cursor Pro; the rest banned AI.
Developers expected a 24% speedup using AI. Instead, completion times rose by 19%. 
METR researchers explained:
“Surprisingly, we find that allowing AI actually increases completion time by 19%— developers are slower when using AI tooling.”
Only 56% had prior experience with Cursor. Most (94%) had used some web-based large language model tools before. Researchers trained everyone on Cursor ahead of the study.
The slowdown may come from the extra time spent prompting AI and waiting for responses. Plus, AI still struggles with large, complex codebases like those in this trial.
METR stressed these results aren’t definitive and don’t dismiss other studies showing AI tools can speed up workflows. They also expect AI improvements soon could reverse this trend.
Still, this adds to concerns about AI coding tools. Other reports have flagged AI-generated bugs and security risks in code.
For more details, check the full METR study.
