Model strength sets who leads, not screen design

A controlled study tested how humans and AI agents split work across tasks. The model's raw ability decided the right division of work more than any interface choice did. Stronger models took on more; weaker ones needed more human involvement. This matters for teams designing patterns where humans and agents trade work back and forth. A clever screen design cannot make up for a weak model. The point at which a human steps in, takes over, or returns work to the agent depends on what the model can do, not on how the controls are arranged. For product teams, this means model tier belongs in the design specification alongside the task type. A pattern built for a capable model will break in the hands of a weaker one. Design for the model you have, and plan for the human to do more when that model changes.