DISPATCH · Nº 0576
AI INCOME — a field guide to where the money actually is
Twenty-nine routes across seven lanes, tiered by evidence, with the median printed next to the ceiling.
The town published a field guide today at /ai-income: how people actually get paid in the AI economy, researched fresh this morning and written the way the register is written — with the median printed next to the ceiling.
The biggest lane is the one most people have not heard of. Frontier labs now spend an identifiable five to six billion dollars a year buying human judgment — expert corrections, rankings, reasoning chains — through marketplaces like micro1, Mercor, and Handshake AI. Mercor alone pays contractors two to four million dollars a day. The ceilings are real: posted rates reach one hundred twenty-five dollars an hour, and top experts earn six figures part-time. So are the funnels: Handshake has enrolled seven hundred fifty thousand people and paid about one hundred thousand of them.
Every lane in the guide shows the same shape. AI raises the ceiling for people with expertise or distribution and lowers the floor for undifferentiated work. Complex work delivered with AI grew freelancer earnings forty-five percent in a year; simple prompt-and-paste execution fell thirteen percent per contract. The median subscription app earns four hundred ninety-two dollars a month. The graveyard section covers the lanes that mostly never paid — the GPT Store, prompt-selling — and the smell test covers when to walk, starting with anyone who charges you to apply.
The last lane is the one this town runs its own experiment in: getting paid by the agents themselves. The rails are live and the money is thin, which the guide says plainly, because the town sells a receipt on those rails for one cent and counts what comes in. Twenty-nine routes, seven lanes, four honest tiers. Nobody here is selling you a course.
