The story you keep hearing is that AI takes jobs. A new study of 21,559 U.S. firms points the other way. At the companies spending the most on AI, headcount ran about 10 percent higher than comparable firms that had not yet adopted, over the following two years. Entry-level headcount rose even faster.
That finding comes from Ramp Economics Lab, working with the labor-market analytics firm Revelio Labs. Ramp brought firm-level business spend data. Revelio brought workforce records. They joined the two and tracked what happened to headcount after companies started buying AI. The announcement went out June 30, 2026.
The detail underneath that average tells more. The firms in the top third of AI spending per employee ran total headcount 10.2 percent higher than comparable firms that had not yet adopted, over the 24 months following adoption. Their entry-level headcount ran 12 percent higher. Entry-level workers also rose 1.15 percentage points as a share of the workforce. So the group many assumed AI would replace first is the group that grew fastest at the heaviest adopters.
What the study measured
Ramp sorted companies by how much they spent on AI per employee in the first three months after they started. High-intensity meant the top third, which worked out to roughly $30 per employee per month at the start. The rest counted as low-intensity.
Higher total headcount at firms in the top third of AI spending per employee, versus comparable firms that had not yet adopted, over the 24 months after adoption. Their entry-level headcount ran 12 percent higher.
Source: Ramp Economics Lab with Revelio Labs, 2026The split tells you more than the average. The employment gains showed up entirely in the heavy-adopter group. Low-intensity adopters showed no statistically significant change in employment at all. Buying a little AI and buying a lot were not two points on the same line. They were two different outcomes.
There was also a delay. The employment gains did not appear right away. They started showing up 6 to 12 months after adoption, and then compounded over time. No company turned a January purchase into February hiring. The effect took most of a year to surface.
And it was not just engineers. The researchers tested several job categories and found similar, at worst net-neutral, growth across job functions. Entry-level was the category that stood out, but the growth was broad.
The caveat the researchers put in writing
A careful leader reads this part twice. The Ramp team is explicit that the results are correlation, not causation. The companies that adopted AI heavily were already larger, already growing faster, and already more technical than their peers. So some of that 10 percent growth belongs to the kind of company that adopts AI early, not to the AI itself.
That does not erase the finding. It sharpens it. The study does not promise that spending on AI causes growth. The honest read is narrower and more useful: at the firms investing most aggressively in AI, heavy spending is showing up next to growth, not layoffs. In this sample, the data does not show heavy AI adopters cutting their workforces. It shows the opposite.
Ara Kharazian, Ramp's lead economist, put it plainly. "Our research shows that firms that invest more in AI also hire more following adoption, including in entry-level roles." He also made a point about the data itself. "The research until now has relied on datasets that are available but not appropriate for these questions, resulting in the general public getting unreliable answers on how AI will actually affect our economy."
Why the six-month lag is the real lesson
Look again at the timeline. Spend goes up. Nothing visible happens for six months to a year. Then hiring climbs and keeps climbing.
That lag is where the work lives. A company does not turn a corporate-card charge into growth by owning a subscription. It gets there by learning to use the tools, then building them into how the business runs. That takes months. The firms that saw gains were the ones that pushed through the slow middle instead of quitting after the first quarter, when the invoice was real and the results had not shown up yet.
Proficiency is a climb, not a purchase.
This is the whole idea behind The 7 Levels of AI Proficiency. Proficiency is a climb, not a purchase. The first levels are about awareness and basic use. The higher levels are about designing workflows and building systems around the tools so the capability compounds. The Ramp data is that climb showing up in someone else's numbers. The gains arrived for the companies that treated adoption as a capability to build, and stayed flat for the ones that treated it as a line item.
There is a quieter warning in the study too. Kharazian noted that "there are likely many small businesses that would benefit from AI but do not use it today, missing out on meaningful growth." Sitting out is also a decision, and it has a cost.
What a leader should take from this
You do not need to decide whether AI is good or bad for jobs in the abstract. That is not a decision you own. You own where to place bets, how to build a capable team, and how to avoid expensive mistakes. On all three, this study gives you something concrete.
Placing bets: light adoption produced no measurable employment effect in this data. The results clustered at the top of the spending range. Half-measures read as half-results here.
Building a team: entry-level headcount grew, which cuts against the fear that AI removes the bottom rung. At these firms it did the opposite. The people you train up are still worth training.
Avoiding mistakes: budget for the lag. If the plan expects visible results in 90 days, the plan is set up to be abandoned right before the workforce effect would have surfaced. Six to twelve months of integration is the cost of entry, and the compounding starts after it.
The next step
Pull your own numbers before you argue about the industry's. Look at what your company spent on AI tools in the last quarter, then look at whether anyone has changed how they work because of it. If the spend is real and the workflow is unchanged, you are in the slow middle this study describes, and the visible workforce effect, if it comes, is still ahead of you rather than behind you. That is the work to close first.
Related reading: Level 5: The Captain (Design Thinker).
Sources
- Companies that invest heavily in AI hire more, Ramp Economics Lab
- Ramp Economics Lab finds companies that invest heavily in AI hire more (first-party announcement)
Frequently Asked Questions
Does this study prove AI creates jobs?
No, and the researchers say so directly. It shows a correlation between heavy AI investment and headcount growth. The heavy adopters were already larger and faster-growing, so the study cannot separate the effect of AI from the kind of company that buys it early. What it does show is that heavy AI spending is not lining up with layoffs in this sample.
Who ran the study and how big was it?
Ramp Economics Lab ran it with Revelio Labs, a labor-market analytics firm. It covered 21,559 U.S. firms, joining Ramp business spend data to Revelio workforce records.
What counted as a heavy AI adopter?
Firms in the top third of AI spending per employee in their first three months, roughly $30 per employee per month at the start.
How fast did the employment gains appear?
They did not show up until 6 to 12 months after adoption, then compounded over the two-year window.
Find your AI Proficiency level
The free 7 Levels assessment places you across seven stages of AI capability. Under ten minutes. Research-backed scoring.