A study of 21,559 U.S. companies from Ramp and Revelio Labs found that heavy AI spenders grew headcount about 10% over two years, with entry-level hiring up around 12%, while modest adopters saw no real change. The dividing line isn’t whether a company uses AI, it’s how deeply it implements it. Access to a capable model stopped being the hard part; the scarce skill is the judgment to redesign how work actually gets done. The model race is becoming an implementation race, and the companies going deep are growing and hiring faster than the ones making shallow bets.
For two years now the story has been the same: AI is coming for your job. Another round of layoffs, another headline blaming the algorithm. There’s truth in some of it, some roles will get automated. But a new study makes me think we’ve been asking the wrong question.
The real split isn’t whether a company uses AI. It’s how far they take it.
Ramp and Revelio Labs pulled together a paper called A New Look at AI’s Impact on Jobs: Firm-Level AI Spending and Workforce Adjustment. They looked at more than 21,500 U.S. companies, matching actual AI spending against real workforce data. No surveys, no guessing at which jobs are “AI exposed.” Just spending, then hiring.
The pattern held up.
Companies with modest AI spending saw no real change in employment. But the heaviest spenders grew their headcount by about 10% over two years. Entry-level hiring rose even more, around 12%.
And it wasn’t just engineering. Sales, customer service, finance, admin, all of it grew. Implementation didn’t shrink these companies. It gave them more room to expand.
The bottleneck isn’t the model anymore
I’ve been saying this all year. Access to a capable model stopped being the hard part a while ago. OpenAI, Anthropic, Google, Microsoft, they’re all shipping frontier-level performance now.
What’s scarce is the judgment to use it well. Which workflows to redesign. Which processes to kill. Which people to augment rather than replace. Which systems to connect. Which governance to put around all of it.
Buying seats isn’t a strategy. Redesigning how the work gets done is. Ramp’s data is some of the first large-scale evidence that the companies doing that hard part aren’t cutting staff. They’re growing.
Other research points the same direction
A Stanford and MIT study of over 5,000 customer service agents found generative AI lifted productivity by around 15%, with the biggest gains going to the newest, least experienced workers. AI acted more like a coach for junior staff than a replacement for anyone.
Anthropic’s Economic Index, built from real Claude conversations, found something similar: 57% of usage looks like people working alongside AI, versus 43% where AI is largely running the task on its own. And Goldman Sachs, looking at the macro picture, still can’t find a statistically significant link between AI adoption and unemployment.
Put it together and the evidence leans toward AI changing work faster than it’s eliminating it.
The caveat that matters
The companies in the Ramp study weren’t average. They were already bigger, growing faster, more technical, better paying, more likely to be venture-backed. The researchers were careful about this. Instead of comparing AI adopters to companies that never adopted, they compared early adopters to similar companies that adopted later. That’s a much more honest comparison than the occupational exposure estimates we’ve been getting for two years.
Worth noting too: most of the early hiring gains are concentrated in tech-heavy sectors, Information, Finance, Professional Services. I wouldn’t assume every industry sees the same curve.
Implementation is the moat now
What stood out to me is that the companies pulling ahead weren’t dabbling. They were spending seriously and sustaining it. That tracks with what I’m watching across enterprise software right now: Microsoft, Google, Salesforce, OpenAI, Anthropic, and most of the big consulting firms are all racing to build out implementation practices. The question in every boardroom has shifted from “should we use AI” to “how do we rebuild the business around it.”
The model race is turning into an implementation race. I think that’s the more interesting race to watch.
So the real question isn’t whether AI will cost jobs. Some jobs, over time, it probably will. The question is whether your organization can actually change how work gets done, not just install another chatbot on top of the old process.
The companies making shallow bets are getting shallow returns. The ones going deep are growing, and hiring, faster than the rest. I think that gap only widens from here.
Sources
- Ramp & Revelio Labs (primary source) — A New Look at AI’s Impact on Jobs: Firm-Level AI Spending and Workforce Adjustment (June 30, 2026). Core source for the 21,559 companies analyzed, 10.2% employment growth among high-intensity adopters, and 12% rise in entry-level hiring. Full research paper.
- Generative AI at Work — Brynjolfsson, Li & Raymond, Quarterly Journal of Economics (also as NBER working paper PDF). ~15% productivity gain, largest for less experienced workers; AI as a coaching and knowledge-sharing tool.
- Anthropic Economic Research — Labor Market Impacts of AI: A New Measure and Early Evidence. Augmentation versus automation and real workplace AI usage.
- Revelio Labs — How AI Is Altering Jobs, Not Eliminating Them. AI adoption correlates with stronger hiring, and augments workers more than it replaces them.
- Ramp Leading Indicators — AI’s First Substitution: Freelancers. One of AI’s earliest measurable effects is reducing outsourced contract work before affecting full-time employment.
- Business Insider coverage — Is AI causing layoffs? This report says it’s complicated. Independent summary of the Ramp and Revelio findings.
- Financial Times — Heavy corporate AI spenders add staff faster than peers. Independent interpretation with caveats on sector concentration and selection effects.
Frequently Asked Questions
Does AI reduce jobs or increase them?
The best firm-level evidence so far points to increases, but only for companies that adopt AI deeply. In the Ramp and Revelio Labs study of 21,559 U.S. companies, high-intensity AI adopters grew headcount about 10% over two years, with entry-level hiring up roughly 12%, while modest adopters saw no meaningful change. AI is changing how work gets done faster than it is eliminating jobs, though some roles will still be automated over time.
Why do heavy AI spenders hire more instead of cutting staff?
Because the value comes from redesigning workflows, not from buying software seats. Companies that implement AI seriously use it to expand what they can do, which creates demand across sales, customer service, finance, and administration, not just engineering. The hard, scarce work is the judgment to decide which processes to change, which people to augment, and which systems to connect.
Does the study prove AI causes hiring growth?
Not on its own. The heaviest AI spenders were already bigger, faster-growing, better-paying, and more likely to be venture-backed, so selection effects are real. The researchers addressed this by comparing early adopters to similar companies that adopted later, which is a more honest comparison than occupational exposure estimates. Most of the early hiring gains are also concentrated in tech-heavy sectors like Information, Finance, and Professional Services, so the pattern may not hold in every industry.
Who benefits most from AI in the workplace?
Less experienced workers, according to multiple studies. The Stanford and MIT research on customer service agents found generative AI raised productivity by around 15%, with the largest gains going to the newest, least experienced staff. In practice AI often behaves more like a coach that spreads the best practices of top performers than a replacement for any single worker.
What does “implementation is the moat” mean for AI?
It means the competitive advantage has moved from access to a capable model to the ability to rebuild a business around one. Frontier-level models are now widely available from OpenAI, Anthropic, Google, and Microsoft, so the model itself is no longer the differentiator. The organizations that win are the ones that can change how work actually gets done, rather than installing another chatbot on top of an old process.