Work
2026
The Farm
Agents learn to climb a sheer wall, fall and begin again, while a counter records every trial.
Across 108 screens, reinforcement-learning agents rehearse the same climb millions of times until a behaviour holds. The training is live, and the count never stops.
The work sets this training beside the human labour it resembles and relies on: migrant construction work, and the outsourced data work that trains AI. In each case the rehearsal is extracted and the one who rehearses is kept out of view.
