Mike McClary used to spend weeks hunting suppliers for a single product. Now he gets cost estimates in minutes—and watches his manufacturing bill drop from $17 per unit to $2.50.
That 85% reduction is the kind of number that makes economists nervous. It proves something is happening to American small businesses right now, in 2026, not some hypothetical future when AI "takes jobs." McClary runs his outdoor gear brand from an Illinois living room. He pulled the Guardian LTE flashlight—a bestseller he discontinued in 2017—off the shelf last year after customers kept emailing asking where to buy it. When he decided to bring it back, he didn't spend days scrolling supplier listings. He opened Accio, Alibaba's AI sourcing tool, described his original design, and received a revised spec within seconds: smaller body, battery-powered charging instead of USB, and a manufacturer in Ningbo, China willing to make each unit for $2.50. The new Guardian flashlight hit Amazon within a month.
"These tools are making sourcing accessible to people who never had access," McClary told MIT Technology Review.
Accio launched in 2024 and crossed 10 million monthly active users in March 2026—roughly one in five Alibaba.com visitors now consults AI before placing a bulk order. For small merchants, this compresses months of supplier research into hours. The traditional workflow required days or weeks comparing factories, negotiating minimum orders, requesting samples. Now AI handles the matching, leaving business owners to do what machines still cannot: relationship management and final negotiation.
This is disruption happening at ground level, in real dollars and real decisions. Yet economists studying AI's workforce impact say they lack the basic tools to measure it.
Alex Imas, a researcher at the University of Chicago, spent years refining predictions about automation's economic effects. His verdict now: the field's standard models are "pretty abysmal." The US government catalogued thousands of individual work tasks beginning in 1998, and researchers at OpenAI and Anthropic have used this taxonomy to rank how "exposed" different jobs are to AI. A real estate agent, by one calculation, is 28% exposed. But Imas calls this framing misleading. "Exposure alone is a completely meaningless tool for predicting displacement," he told MIT Technology Review. Knowing that AI *could* do a task tells you nothing about whether it *will*—or whether it costs less to deploy than the human doing it now.
The tension is stark. McClary's 85% cost cut is measurable, undeniable, happening now. Meanwhile, economists cannot agree on how many people will lose their jobs, or when, or which ones. They can count that AI tools have 10 million users. They cannot tell you what that means for employment.
What would help, Imas argues, is task-level employment data—tracking not just who is employed, but which specific tasks they perform and how that changes over time. Without it, policymakers are flying blind. With McClary, they can see exactly one data point: his costs dropped 85% in a single product cycle.
Both facts are true simultaneously. The disruption is real. The measurement is not.