A research team studying its own AI model-building process reviewed 769 task logs and found that AI agents could account for as much as 55% of method proposals, while more than 85% of final decisions were still made by humans. The study also said roughly one-third of tasks would not have been attempted at all without AI support. The authors cautioned against reading higher agent activity as evidence of stronger autonomy. According to the findings cited by The Decoder, the increase in AI agent participation did not translate into a comparable shift in decision-making authority. Instead, humans remained firmly in control of final calls even as agents contributed a large share of suggestions during the workflow.