Research

The METR Study One Year Later: When AI Actually Slows Developers

In early 2025, METR (Model Evaluation and Transparency Research) ran a randomized controlled trial that caught the industry off guard. Experienced open-source developers—people with years on mature, high-star repositories—were randomly assigned to complete real tasks either with AI tools (Cursor Pro with Claude) or without. The result: with AI, they took 19% longer to finish. Yet before the trial they expected AI to make them about 24% faster, and after it they believed they’d been about 20% faster. A 39-point gap between perception and reality.

The AI Productivity Paradox: Why Experienced Developers Are Slowing Down

There’s something strange happening in software development right now, and I think we need to talk about it.

Recent research has surfaced a troubling finding: experienced developers working on complex systems are actually 19% slower when using AI coding tools—despite perceiving themselves as working faster. This isn’t a minor discrepancy. It’s a fundamental disconnect between how productive we feel and how productive we actually are.

As someone who’s been experimenting with AI tools extensively (and writing about the results), this finding resonates with my experience. Let me break down what’s happening and what it means for engineering teams.