Dear Colleagues: The most important economic question of the next decade is how AI will affect jobs. Everyone wants to write that paper. Until now, no one has had the right dataset, so existing research has relied on a combination of guesses, surveys, AI exposure scores, and self-interested punditry. In fact, a recent paper from Stanford said the ideal dataset would use “business spend data” and cited Ramp data specifically.

Today, we published our first working paper: A New Look at AI’s Impact on Jobs. It uses firm-level spend data from Ramp joined with workforce data collected by Revelio Labs. In a sample covering more than 21,000 U.S. firms, we find that companies that invest heavily in AI grow headcount 10% over the two years following adoption. Entry-level headcount grows 12%.

Read the full paper and explore the data from Ramp Economics Lab here. The rest of this letter will outline our key findings, plans for future research, and some of my thoughts on what that means for our economy.

• Firms that adopt AI grow headcount 10.2% over the two years following adoption, but these gains are entirely driven by high-intensity adopters. Low-intensity adopters see no statistically significant change.
• Entry-level headcount grew even faster. At the companies making the largest AI investments, entry-level headcount grew 12% over the two years following adoption.
• AI adoption and the associated gains are unevenly distributed. AI adopters are already larger, more engineering-intensive, more likely to be venture-backed, and faster-growing than non-adopters. These firms then grow faster upon adoption.

Our main result says yes. Companies that use AI grow their headcount 10% over the two years following AI adoption. Two caveats to that:

• AI adoption is subject to a minimum threshold: You only see gains when you adopt AI with “high-intensity.” In our analysis, we defined high-intensity as firms in the top third of per-employee-per-month spend on AI in the first three months. That’s not necessarily the ideal way to describe great AI adoption, but it correlates with what I wanted to capture: these are the firms that spend on multiple models and use the most advanced and productivity-enhancing products available (coding agents and APIs as opposed to simple chat subscriptions). Note the top-third isn’t spending that much. About $30 per employee per month in the first three months, rising thereafter, but not thousands of dollars.
• The gains are subject to a learning curve: Firms don’t increase their headcount until 6-12 months following adoption, and these gains compound over time. It takes time for best practices to proliferate through teams and individual and organizational workflows, before they change company operations enough to drive growth and affect hiring.

After we answered the first question, whether firms that adopted AI were hiring more, our paper shifted to answer: are they hiring differently? We tested several job categories and education-levels. Broadly, we found similar (at worst, net-neutral) growth across job functions, except for entry-level positions.

Entry-level hiring grew 12% in the high-intensity group. It grew so fast that at the end of the 24-month analysis period, firms in the high-intensity group also increased their workforce share of entry-level workers (up 1.15 percentage points compared to the control group).

It’s the first and only example we’ve seen so far that high-intensity AI firms are selecting different kinds of candidates. In this case, we believe they are selecting for a new set of skills, specifically, people who know how to use AI and use it well. Entry-level workers, especially recent graduates and college students, are a natural place to look.

These results are early, and we intend to track and update them over time, but it’s a promising (and intuitive) result for young people who have gotten seemingly exclusively negative news about their job prospects in recent years. I have seen it firsthand from hiring managers we spoke to in developing early versions of this paper.

A side-result of this paper is that we found that AI proliferates through networks, and those networks can lead to outcomes that are economically suboptimal.

For example, who funded you is a better predictor of AI adoption than the sector you’re in. VC-backed companies of any kind are more likely to use AI, and use it intensely, than legacy tech companies. There’s also proliferation around who you know and the places you hire from. California-based tech companies are more likely to use AI than similar companies in New York. There’s no inherent reason for that, other than the fact AI spreads through networks.

Consider small businesses for example. They are less likely to use AI in the first place, but when they do, they use it more intensely. That makes sense to me: A small firm may not have a dedicated engineering or finance team. If AI lowers the fixed cost of building software, handling administrative work, doing analysis, or improving customer support, the gains can drive outsized growth and unlock new revenue streams that previously required higher fixed costs in the form of new salaries. Ultimately, I think this is the mechanism by which firms increase headcount upon AI adoption, even though AI presumably fulfills new and existing tasks for their teams. They can go do more things now.

If you are a young person entering the labor market and choosing between two otherwise similar firms, choose the one that’s using AI. It’s more likely to grow faster.

If you are an engineer worried that AI will eliminate engineering jobs, our evidence says AI adopters are hiring engineers faster, not slower. Learn the latest technology to best position yourself for high-intensity AI adopting firms.

If you are a business owner that’s tried AI and didn’t see the gains, keep at it. Our results show the gains from AI are subject to a learning curve and a minimum threshold of adoption. Consider your experience quantitatively common but not final. The firms who use AI well have no incentive to publish their playbook, so it’s hard to learn best practices without direct experimentation.

Finally, if you are reading headlines where CEOs blame layoffs on AI, be skeptical. This work is not final, and we intend to update these results as more data comes in, as well as testing more outcomes to see how hiring is changing. This will not be my last paper, but my hope is that readers – including workers, business owners, and policymakers – can use today’s results to better their own decisionmaking and get better outcomes.