Track what is changing.

Approval fatigue, deskilling and responsibility without control: supervising AI agents is a work activity with its own psychosocial risks.

Expertise does not necessarily disappear: it shifts from detailed code production towards architecture, evaluation and supervision of the outcome.

Threatened expertise, partial autonomy, conflicts of values, intensity and a high cost of leaving: five tensions in a combination of risks that remains poorly documented.

Faster agents can also accelerate the pace at which humans must decide, supervise and respond, intensifying hyperconnectivity and difficulty disengaging.

Five workers were recruited, scheduled and managed by AI agents, shifting the issue from algorithms operating in the background to AI placed directly in authority.

Ten results produced by an internal model on open research problems illustrate the shift from AI as a researcher’s assistant to AI as a direct producer of knowledge.

Task crossover may broaden autonomy while changing workload, teams, professional values and job security.

A mathematician’s account reveals tensions involving meaning, recognition and expertise before employment disappears.

Effects differ across tasks and workers; uniform expectations can move pressure onto workers instead of redesigning deployment.

AI capabilities may be advancing faster than political institutions, requiring anticipation of effects on security, employment, taxation and civil liberties.

If AI largely replaces human labour, individuals could lose political and social bargaining power while capital and power concentrate.