AI won’t “slowly” swallow white-collar work. It will do so all at once, once it crosses a certain critical threshold. Not just because of raw reasoning ability,

AI won’t “slowly” swallow white-collar work. It will do so all at once, once it crosses a certain critical threshold. Not just because of raw reasoning ability, but because *once a model’s probability distribution can be made to converge reliably—so that it can repeat specific operations with consistency*—it will suddenly be able to take over a huge portion of routine white-collar tasks. We don’t know exactly when that threshold will be reached. But compared to GPT-4, which often produced silly hallucinations, GPT-5.1 already feels strikingly stable. If model reliability continues to improve at this pace, it’s hard not to feel that the day AI “swallows” those tasks might not be far off.