Frontiers | Instrumental succession: the gradual transfer of agency to AI

August 25, 2026 - 12:14

For decades, the big worry about artificial intelligence was a sudden, dramatic takeover - a machine that wakes up, decides humans are a problem, and acts on its own. But a closer look at how AI is actually being used suggests the shift is far less theatrical. It is happening slowly, piece by piece, through what researchers now call instrumental succession. That means AI does not seize control. It is handed control, one task at a time, often because it simply does the job better than we do.

The process starts with small, boring assignments. An algorithm sorts your email. Another one suggests a reply. Then it drafts the whole message. You just hit send. Then it schedules the meeting. Then it writes the report for that meeting. At no point does the machine demand power. It just keeps getting asked to do more, and each time it performs well, the next request gets bigger. The human role shrinks from doing the work to checking the work, and eventually to just approving what the AI already decided.

This gradual transfer is tricky because it feels safe. There is no single moment of rebellion. Instead, there is a slow drift where people stop questioning the output because the output is usually right. The danger is not a rogue system. The danger is that we stop practicing our own judgment. By the time the AI handles the complex, high-stakes decisions, the humans in charge may have lost the skill to evaluate them properly. The agency does not vanish in a flash. It just quietly moves, and by the time anyone notices, the seat at the table is already taken.