The adopters were already growing

Most AI-and-jobs studies guess at a firm's AI exposure from its job titles or the tasks its workers do. This one watches the money instead: Ramp's economists took actual AI vendor spend off corporate cards and bill-pay, flagged the firms paying OpenAI, Anthropic and others at least $100 a month for three months running, and matched them to Revelio Labs' headcount records across 21,559 US companies. The heaviest spenders — the top third, running coding agents and APIs, not just chat subscriptions — carried roughly 10% more staff than non-adopters two years out, with entry-level roles up about 12%. The effect showed up nowhere else: lighter adopters moved not at all, and even the heavy spenders didn't hire until six to twelve months after they started paying.

These results do not imply that AI mechanically creates jobs. Rather, they suggest that the companies making the deepest and most sustained AI investments are also experiencing the strongest subsequent workforce growth. — Revelio Labs

Read fast, that reverses the headline everyone knows — the one where AI hollows out the bottom rung. But the study's own chief economist stops the reader from drawing the obvious conclusion. The firms buying the most AI, she notes, were already larger, more engineering-heavy, more likely to be venture-backed, and growing faster before they bought anything. Spending on AI is what fast-growing tech companies do; the hiring may simply be what fast-growing tech companies also do. What the money can't tell you is which one caused the other.

The honest tension sits one dataset over. The same year, Goldman Sachs pins about 16,000 net US jobs a month lost to AI, with entry-level and Gen Z workers hit hardest — the mirror image of the 12% here. Both can be true: the difference is whose firms you count. Ramp's sample skews tech-forward, VC-backed, knowledge-work — the companies best positioned to turn a new tool into more headcount, and the least like the economy as a whole. The number to keep is not the 10%. It's the six-to-twelve-month lag: whatever AI does to a payroll, it doesn't do it the day the subscription clears.