Hayden Brown runs Upwork, the freelancing platform where 18 million independent workers find clients. That gives her an early view of hiring trends, months before they show up in official jobs numbers. Her latest read on the AI jobs story cuts against both extremes people usually hear.
The most striking number: nearly a quarter of Upwork's business clients that shifted work over to AI have already moved it back to human freelancers, or are planning to. In Upwork's own testing, AI agents left to deliver finished work with no human involved almost never complete even simple tasks successfully. Add a human expert to check and refine the work, and the success rate jumps sharply.
This lines up with independent research. A recent benchmark from UC Berkeley tested leading AI systems against nearly 1,500 real work assignments gathered from working professionals across 55 industries, covering tasks like legal filings and financial models, and the AI completed only about one in four of them correctly. A separate study found the best performing AI model still only succeeded on roughly a quarter of tasks, with weaker models succeeding on less than one in ten.
None of this means AI is useless. It means the current generation of AI works best as an assistant that speeds up a skilled person, not as a replacement that runs unsupervised. Businesses that skip the human check are the ones ending up back where they started, minus the time and money spent finding that out.
There is a second layer to this story worth watching closely: some layoffs blamed on AI may not really be about AI at all. When companies like Wix cut large portions of staff and point to AI as the reason, researchers who study corporate layoffs note this is a familiar pattern. Blaming a new technology for job cuts is a strategy companies have used for decades to explain restructuring, and it tends to be received well by investors.
Meanwhile, freelancing itself is growing fast, and not just because people are being pushed out of full-time jobs. Freelancing among US knowledge workers jumped by ten percentage points in a single year. Younger workers are driving much of this, partly because freelancing lets them build AI skills through real client work rather than a training course, and those skills come with a real pay bump.
For any business leader, the takeaway is simple. Treat AI as a productivity boost for skilled people, not a way to remove people from the process entirely. Companies that made that mistake are already paying to reverse it.