AI's first substitution: freelancers

Everyone wants to know if AI is replacing workers. The problem is measuring it. The research so far has tracked AI-exposed job titles and their respective hiring trends, but the definition of “AI-exposed” shifts every time we get a new, more capable model.

The ideal dataset would track firm-level AI adoption alongside hiring, but the labor market is noisy. If you look at the aggregate workforce — nurses, chefs, construction workers — you aren’t going to see the impact for years. It’s too diluted.

So where do you look to see the future? Economists look at the margin, the people likely to be affected first. Here, that’s workers with the least friction: freelancers. Why?

1. Freelancers are heavily exposed to economic transformations already. No severance. No HR process. Companies can let contracts expire with little penalty and stop posting new ones.
2. Freelance work is usually defined by tasks, not roles. Writing a blog post, designing a logo, fixing code. These are well-scoped tasks that don’t require a lot of institutional knowledge. And LLMs are really good at them.

That’s the focus of a new paper from Ramp Economics Lab by my colleague Ryan Stevens, Ramp’s director of applied science. He tracked firm-level spending on freelance marketplaces (Upwork, Fiverr) and AI model providers (OpenAI, Anthropic) from 2021 to 2025.

Two findings stand out.

First, businesses are shifting spend from freelancers to AI. The share of total spend going to labor marketplaces fell from 0.66% in Q4 2021 to 0.14% in Q3 2025. AI model provider share for the same companies rose from zero to nearly 3% over the same period. More than half of the businesses using freelancers in 2022 have stopped entirely. And the companies that used to spend the most on freelancers shifted to AI the fastest.

Second, AI is dramatically cheaper. Firms most exposed to AI, those that spent the most on freelancers pre-ChatGPT, substituted at a rate of about $1 in reduced freelance spend for $0.03 in AI spend. That's roughly 25x cost savings. Even the middle-exposure group showed that for every dollar saved on freelance labor, they spent $0.30 on AI. The true number is probably somewhere in between, but either way: AI is not a 1-for-1 replacement. It's an order of magnitude cheaper and companies are keeping or re-investing the savings.

These results suggest a blindspot for policymakers. While Americans are concerned that AI will cause permanent job loss, freelancers are in a more precarious position. They lack the protections that come with traditional employment (benefits, unemployment insurance). They're also outside the standard policy toolkit for managing labor market disruptions. If they're first in line for displacement, what happens to them?

There's an irony here. A few years ago, the discourse worried about gig work eroding traditional employment. Uber and Upwork were threats to the social contract. Now the gig economy may be the first victim to automation. Rideshare drivers from self-driving cars. Freelancers to Claude and ChatGPT. The narrative flipped fast.

None of this tells us what aggregate employment will look like in 10 years. Micro substitution doesn't imply macro job loss. AI will create more roles than it destroys, but this first result tells us where the displacement is happening first.

I highly encourage you to read Ryan’s full paper for the complete methodology, with an extensive description of our approach and analysis. This is an area of emerging research — and an important one — where Ramp data will be particularly critical in tracking firm decision-making.