The research, conducted by the non-profit Model Evaluation & Threat Research (METR), set out to measure the real-world impact of advanced AI tools on software development.