A new study from Ramp and Revelio Labs offers valuable evidence about how companies behave after investing heavily in artificial intelligence. Unfortunately, the resulting headlines, hyped by the authors themselves claiming AI doesn’t cause job loss, push the findings past their breaking point.

The defensible conclusion concerns AI job growth inside a specific group of early-adopting companies. Firms that spent heavily on AI subsequently expanded their workforces faster than comparable companies that adopted the technology later. That finding deserves attention from business leaders and policymakers.

However, the study can’t tell us whether AI displaces jobs. It measures employment growth inside adopting companies, while displacement can occur through layoffs, reduced hiring, fewer contracts, or job losses at competitors. Policymakers and business leaders should treat the research as evidence about the growth of AI-intensive firms, rather than as a verdict on the future of work.

The researchers combined Ramp transaction records with Revelio Labs workforce estimates for 21,559 U.S. companies. They classified companies as adopters after at least three consecutive months of qualifying AI spending, then divided them by spending per employee.

The underlying AI adoption data show that companies in the highest spending group had about 10.2% more employees during the first 24 months following adoption than companies that adopted later at a similar intensity. Their entry-level head count rose about 12%. Companies making lower-intensity investments recorded no statistically significant increase.

Companies that invest seriously in AI appear to grow faster.

The reasonable conclusion follows directly from those numbers: Companies that invest seriously in AI appear to grow faster. Buying a few chatbot subscriptions showed little measurable employment effect. Sustained spending on coding tools, application programming interfaces, and other AI systems correlated with broader hiring across engineering, sales, administration, and customer service.

That pattern fits established economic logic. Technology that lowers production costs can increase output, improve service, and help companies reach more customers. Research on AI productivity gains has already shown that generative AI can help customer-support agents complete more work, with especially large benefits for less-experienced employees. Productivity can support hiring when growing demand outweighs the amount of labor saved through automation.

The study also reveals how unevenly those gains appear. Statistically significant sector-level employment growth emerged only in the information sector, which includes many software, internet, and media firms. High-intensity adopters in that sector increased head count by about 13.4%. Results for other broad sectors remained close to zero, statistically uncertain, or both.

That concentration limits broad conclusions. Technology-oriented companies can turn faster software development into new products, lower costs, and additional sales relatively quickly. Hospitals, banks, manufacturers, and government agencies operate under different technical, regulatory, and organizational constraints. Their employment response may look very different.

The companies in the sample also entered the study with unusual characteristics. High-intensity adopters were more heavily concentrated in technology-related industries, paid higher salaries, and grew faster before adopting AI. Their median year-over-year head count growth before adoption reached 8.9%, compared with 1.6% among companies that never adopted during the study period.

The researchers address part of this selection problem by comparing early adopters with companies that adopt later at a similar intensity. That design improves substantially on a simple comparison between adopters and nonadopters. It still can’t fully separate AI’s effects from venture funding, strong management, successful products, or expectations of rapid demand. A company preparing for expansion may increase hiring and AI spending simultaneously.

More fundamentally, employment growth can’t rule out displacement. Consider a company that would have hired 500 workers to meet rising demand without AI. After automating part of the work, it hires 200. Its workforce grows, yet AI reduces potential employment by 300 positions.

The missing positions never register as layoffs. They appear as vacancies never posted, contractors never retained, and departing employees never replaced. Ramp’s separate research on AI displacement risk found that companies with substantial previous spending on freelance platforms adopted AI more aggressively while reducing freelance expenditures. Payroll employment can grow while outside workers lose opportunities.

Firm-level growth also says little about economywide employment. An AI-enabled company might gain market share and add 100 workers while several competitors cut 200 positions. The adopting firm records job growth even as the industry contracts. Policymakers care about both outcomes.

The wider evidence on AI employment effects remains mixed.

The wider evidence on AI employment effects remains mixed. Stanford researchers have reported declining employment among younger workers in highly AI-exposed occupations. Research using Danish administrative data found small labor market effects during the first two years of widespread chatbot use. These studies measure different populations and outcomes, so their findings can coexist.

The Ramp study’s authors complicate the discussion by framing their results as counterevidence to broad job-loss predictions and stating that their preferred estimates show no evidence of AI producing layoffs. That language reaches beyond the outcome they measured. Head count data can’t reveal every layoff when companies hire elsewhere, and it can’t observe avoided hiring or losses outside the firm.

This overstatement deserves criticism, though it shouldn’t eclipse the study’s genuine contribution. Promotional conclusions narrow the AI public debate into a false contest between mass unemployment and universal prosperity. The evidence points toward a more complicated transition in which some companies grow, some workers gain leverage, and others lose opportunities.

For Washington, the policy lesson centers on measurement. Effective AI workforce policy requires data on layoffs, vacancies, wages, hours, contractor spending, output, and employment at competing firms. Aggregate head count at successful adopters captures only one part of the picture.

The Census Bureau’s work on nationally representative business AI adoption provides a better foundation for tracking differences across industries and company sizes. Congress should support efforts to connect adoption data with administrative records while protecting business and worker privacy.

The Ramp study offers encouraging evidence that intensive AI investment accompanies faster corporate growth, especially among technology-oriented companies. That finding could matter greatly for productivity, competition, and economic policy. It leaves the displacement question open, where the evidence says it belongs.