For about two weeks, Carolina Perez Sands — a speech and language pathologist based in Brazil — worked as a contractor for Mercor, an AI training firm, teaching a language model how Brazilian Portuguese sentences should 'breathe,' how a piece of creative writing should earn its emotional turn rather than announce it.
Each day she sat down with the model's output and marked it. She explained patiently why the phrasing was wrong, why a native ear would flinch. She externalised the clinical intuition she'd built over a decade — the pattern recognition that lives beneath the textbook.
And then one morning she sat down to the same task and found she had nothing to say.
'The model has absorbed me,' she told The Wall Street Journal's podcast in June 2026. 'My corrections became unnecessary within roughly a week.'
Mercor was by then paying more than thirty thousand contractors upward of four million dollars a day to help make their own jobs — and the jobs of their colleagues — obsolete. The postings read like a fever dream of the professional class: $225 an hour for a voice actor with fluent Hebrew, a doctorate-holding physicist in general relativity, physicians who could describe the texture of primary care in Rwanda. Data labelling had moved up the value chain.
Carolina didn't quit because the pay was poor, though for many in the sector it became so. She quit because she understood precisely what her work was. 'My job,' she said, 'was actually making this more of a monster.'
She has since left the industry entirely.