Consider an AI researcher midway through a programme, with experiments running, explanations still tentative and a feel for the laboratory’s instruments. A stronger replacement should take over those responsibilities and bring better reasoning to them.
Two workers given the same summary can still decide differently, so a handover succeeds only if the successor does the right thing when a delayed obligation comes due.
A handover that keeps the abilities
A good handover carries the work and the abilities it depends on. If an ability does not survive the move, the team learns that before it assigns work that relies on it.
The same standard applies whenever a long project changes hands: when context is compressed, a process crashes or one worker passes a task to another. Long projects need those transitions to be routine, with a known cost and nothing lost. In our September study, durable agents ran 14.8 times longer before a person had to rebuild their work.