Most AI tools are quietly degrading the people who use them. A new framework published in UX Collective by designer Daisy Chen argues that every AI product is, intentionally or not, a training system for its users, shaping which skills they practice and which ones atrophy. The piece roots this in Bainbridge's 1983 'Ironies of Automation' research from aviation and nuclear plant ergonomics, and builds a four-part design checklist that product teams can apply directly.
The framework is worth reading in full because the interesting work happens in the middle sections, not the conclusion. Chen breaks task delegation into four cognitive stages borrowed from Parasuraman's automation research, then maps them against four human control levels ranging from manual to full AI execution. The control-level selection criteria are specific: reversibility of errors, time-criticality, and user experience all determine how much authority to hand the AI. The piece also cites John Maeda's Design in Tech Report 2026, which frames the core shift as moving from designing how users execute tasks to designing how users evaluate outputs.
The framework's fourth pillar, designing for coevolution, is where the argument sharpens. Chen is not asking teams to slow AI down arbitrarily. The ask is to build systems where user judgment grows alongside AI capability, not in spite of it. For anyone building agentic tools or copilot features right now, the checklist items are concrete enough to audit against an existing product tomorrow.
[READ ORIGINAL →]