What the jobs report isn't telling you about AI & the workforce, with Upwork’s Hayden Brown

Hayden Brown describes the current labor market as subdued and sluggish across several quarters, consistent with broader jobs data, but argues that traditional monthly reports miss critical dynamics. The most recent jobs report showed 23,000 jobs lost while unemployment simultaneously fell, producing contradictory signals that Brown says illustrate why headline figures are incomplete. Small business customers on Upwork have pulled back hiring across all channels as the macroeconomic slowdown hits them hardest, yet Upwork's AI category grew over 22% year over year, and demand for talent with AI skills is where energy and money is flowing from small businesses through large enterprises.

Freelancers with AI skills earn 34% more per hour than those without, and freelancing among US knowledge workers rose 10 percentage points to 38% of the population, up from 28% just a year ago. Brown attributes faster AI skill adoption among freelancers to the fact that their income depends directly on staying current. Some businesses are also hiring freelancers partly as a backdoor method to upskill their in-house teams on AI, adding a structural dimension to the demand shift that payroll data does not capture.

Brown rejects both the extreme view that AI will eliminate all jobs and the pure optimist view, placing reality in the middle with no mass job destruction visible yet, while noting AI's effect is more destructive on the negatives and more creative on the positives compared to prior technology cycles. He offered as context that 60% of jobs existing today did not exist in 1940. He warns that framing the question as humans versus machines distracts from building a future where humans win with AI rather than at the loss of jobs, opportunities, and purpose.

Brown says many CEOs engage in AI washing by attributing workforce reductions to AI to boost stock price rather than acknowledging prior strategic failures, and that companies which let go of workers citing AI are now finding those workers are still needed because AI cannot do everything they hoped. A recent Upwork survey found 23% of clients have already moved work back to humans from AI or are about to do so. When clients tested AI agents to deliver work end to end, the failure rate was incredibly high even on fairly simple tasks, but adding a small amount of human expertise and a few exchanges with an expert raises agent task success rate by more than 70%.

Upwork cut a quarter of its corporate staff in May, which Brown says was driven by changing market conditions and a goal to expand profitability. He clarified that AI was a contributor to efficiency gains but not the biggest reason for the organizational changes nor the biggest benefit derived from them. The restructuring involved asking what each function would look like built from the ground up using the best available technology, rather than simply automating existing roles.

Upwork launched an MCP server so that AI agents can authenticate with the platform, allowing Upwork to screen good versus bad actors before permitting them to post jobs, make hires, and manage projects. Clients working inside tools like Claude or ChatGPT who hit a capability wall can use the MCP to connect directly to a freelancer in real time. Brown treats ChatGPT, which has over 900 million users, and Claude as new origination and advertising channels for Upwork rather than threats. The platform, which had 18 million freelancers active in the last year, holds what Brown describes as the most verified work history data in the world at scale because it observes people doing actual work with dollars attached, making that data more strategically important as AI-generated spam proposals and application noise grow.

Brown flagged a rising role he calls the AI orchestrator, a generalist who can coordinate AI tools across domains, as increasingly popular on Upwork. On model strategy, Upwork takes a gateway approach, constantly testing multiple models including open source options, and token costs from major US providers have risen enough to make open source attractive. Upwork can achieve 10 times savings or more using open source models in certain use cases, and Brown says open source models from China can be used safely with the right controls and are not categorically inferior. Brown added that even a one-year planning horizon is too far out given how fast the AI landscape is shifting, and that Upwork has shortened planning cycles and moved to ranked priority lists so that lower-ranked work is not pursued until higher-ranked items are satisfied.