Conscripting engineers to make training data won't push AI (opens in new tab)

In spring 2026 Meta reassigned about 6,500 engineers into a unit that manufactures AI training data. My forecasts say the bet holds and does not blow up: the org stays near 4,600 through year-end, the surveillance program survives, and Meta's top model stays about 15 points behind the frontier with no sign of closing the gap in 2026. The talent cost is real but lopsided.