A fresh empirical look at Upwork hiring patterns after ChatGPT's arrival shows clients dialing back their focus on worker credentials, experience, and self-presentation while placing more weight on posted price, with the shift most visible in job categories where AI exposure is high.
New research: Is AI making employers view labor as more of a commodity?
In a large online labor market, we find that post-ChatGPT, clients place less weight on signals of human capital and more on price when hiring.
Demand premiums that once rewarded high-human-capital profiles shrank, pushing hiring toward lower-priced workers and flattening differences that used to matter.
Search and recommendation systems built around skill signals could lose relevance if price becomes the dominant filter, raising open questions about how marketplaces will adjust incentives for workers and clients alike.
Positive users called the research on post-ChatGPT employers prioritizing price over human capital interesting and exciting to read, while negative users viewed the trend as concerning for job seekers.
No Digg Deeper questions have been answered for this story yet.
Some interesting findings suggesting that, by leveling performance, AI also commoditizes contract labor.
New research: Is AI making employers view labor as more of a commodity?
In a large online labor market, we find that post-ChatGPT, clients place less weight on signals of human capital and more on price when hiring.
See full paper with amazing PhD student Niuniu Zhang here: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6968139
and the tl;dr: slides here: https://auyonsiddiq.com/commoditization.html
The motivation is straightforward: Several prior studies have show that AI can compress differences in worker productivity. In a recent theoretical paper, Fukui, Nakamura, and @JonSteinsson suggest that this could commoditize labor. We complement this by providing empirical evidence.
This extends prior work on how AI changes the value of human capital signals (e.g., cheap talk by @bocowgill) , by suggesting even costly signals of human capital (like education) may be changed by AI, because markets anticipate compression in productivity.
We also build on work by @xianghui90 and @oren_reshef that shows compression in worker earnings post-ChatGPT, by probing specifically for possible commoditization.
We use a panel of 50k workers that completed 2.23m contracts on UpWork from 2021-2026. We represent the detailed profiles of workers across different job categories as text embeddings, which are used to predict worker-level demand in the labor market.
We then quantify the importance of different types of human capital information and price (i.e., hourly rate), and ask whether the release of ChatGPT changed the importance of these attributes in the eyes of the demand-side of the market.
Understanding whether AI is commoditizing labor is important for how labor markets are organized, workers' incentives to invest in human capital, and labor welfare more broadly.
Although price becoming more important might hint at commoditization, there may be other explanations, like clients using high prices to infer worker quality.
But! We also see demand move *toward* lower-price workers, which leans more toward commoditization as the explanation.
Cool paper!!! Excited to read, and maybe you show this, but could shift in demand be from shifts in the composition of who is hiring on upwork?
For example, if the firms that wanted the highest human capital labor are now using AI themselves instead of hiring on upwork, then for non-contract labor we might see decommodization as people are augmented by AI instead of outsourcing to a contractor on upwork.
So while upwork workers become commodities, other workers might not?
Inspired by this super important and underrated analysis from @arakharazian https://www.google.com/url?sa=t&source=web&rct=j&opi=89978449&url=https://ramp.com/leading-indicators/ai-labor-market-impact-freelancers&ved=2ahUKEwjl89-i2JWVAxV_v4kEHYivG7wQFnoECB8QAQ&sqi=2&usg=AOvVaw0dE_IqDYj5GWP7EQyMbaK5
@auyonomous Thanks for the shoutout, Auyon. Looks like a really interesting paper, and I look forward to reading it.
@auyonomous very interesting!!
@auyonomous
@Cronesie @auyonomous Exactly my thought, I go to upwork for fairly generic work, even within my industry. I have never been able to use for truly specialist work.
@emollick 100%. We had 1000 invoices to extract some data. We would have paid manual labor to do that. Instead, we use Claude Code parse it.
@bocowgill Thanks for trailblazing in this space 🙏
@auyonomous Aren't the tasks that are amenable to UpWork etc by definition more prone to commoditization than other types of work?
@auyonomous Got my doubts
Thanks! And that's a fantastic point, we will add a comment in the paper.
Changes in the composition of clients is entirely plausible, especially in light of the paper by Demirci, @HannaneJonas, @zhu_xinrong, who find a change in job postings.
So yes, the change in "the market" preferences is quite possibly a change in who participates on the demand side, rather than a fixed set of clients seeing their preferences change over time.
The paper I mentioned: https://pubsonline.informs.org/doi/abs/10.1287/mnsc.2024.05420
@emollick When has contract labor been anything other than a commodity?
@emollick That’s because AI isn’t changing the transfer of context and intent. Without that, every prompt has a rocky foundation and is built on assumptions that compound as soon as the first detail is wrong. You’re scaling your misconceptions and people see that. People use AI all wrong.
