Designing work so that people and AI systems each do the parts they are best at, and hand over to each other cleanly. It treats the AI as a team member with a defined role rather than as a simple tool or a replacement. Getting it right depends as much on who decides what, and when, as on how capable the AI is.
Technical view
Draws on decades of human factors research in aviation, defence and process control, where the older name human-machine teaming is common. Design questions include allocation of function, a shared picture of the situation, how the system signals its confidence and limits, how authority passes between person and machine, and how people keep the skills to take over. Known failure modes are automation bias, complacency, deskilling and badly timed handovers, where someone is asked to intervene without enough context. Evaluation should measure how well the pair performs together rather than the AI alone.
Related
Part of AI Terminology Index, section: Product and Adoption.