METR and Google DORA reports flag AI coding tools slowing devs down
The METR study just dropped a bomb: experienced developers saw a 19% slowdown when using AI coding suggestions. That’s not AI failing outright, METR’s Gogia says, but real-world friction mixing guesswork AI into exact workflows.
Gogia stressed the bigger picture: 
“The 19% slowdown observed among experienced developers is not an indictment of AI as a whole, but a reflection of the real-world friction of integrating probabilistic suggestions into deterministic workflows,” Gogia explained, emphasizing that measurement should include “downstream rework, code churn, and peer review cycles—not just time-to-code.”
The timing syncs with Google’s new 2024 DORA report surveying 39,000+ engineers. Despite 75% saying AI made them feel more productive, data shows every 25% rise in AI use dragged delivery speed by 1.5% and tanked system stability 7.2%. Plus, 39% barely trust AI-generated code.
This clashes hard with past hype. Earlier MIT, Princeton, and U Penn research used data from 4,800 devs at Microsoft, Accenture, and others, claiming GitHub Copilot users did 26% more tasks and finished coding jobs 55.8% faster. But those tests focused on smaller, controlled tasks, unlike the complex real-life dev scenarios METR examined.
