—Getty ImagesSpencer Handley’s online guitar school, Sonora—which claims professional musicians such as Tom Misch and Billy Strings among its students—had been growing steadily for seven years, mostly unaffected by the spread of AI. Then came Claude Opus 4.5. The AI model, released by Anthropic in late November and trained on specialized data that taught it to complete long running agentic tasks, such as software engineering and administrative work, marked a step change in AI capability, according to multiple entrepreneurs and AI researchers who spoke to TIME. Suddenly, Handley found that the AI could replicate the enterprise software he used to run his business. “I was like, ‘oh, the game has changed. You can clone a billion-dollar company’s [software] with no human intervention,’” he says.The improved AI also rendered some of his employees redundant. All but one of his 12-person team of “setters,” who reached out to potential customers, were let go. So were his sales manager, customer onboarding team, and operations staff. The remaining employees oversaw AI agents that wrote marketing copy, followed up with potential customers, and onboarded new students. By April, he had replaced HubSpot, Calendly, Vimeo, and DocuSign with tools customized to his company, saving him roughly $250,000 a year, and centralized all of his customer data so that he could run AI agents on it more easily. His 48-person company was down to 30 employees without losing revenue. “We actually get slightly better results, and we don’t have to worry about managing people,” he says.While attention has focused on mass layoffs at large tech companies, some economists and entrepreneurs believe the most significant AI-driven workplace changes could emerge first in small firms, which can reorganize around new technology more quickly. “AI adoption is faster in smaller firms, including startups,” the Harvard economist David Deming wrote in 2025. If that’s true, splashy headlines about large tech firms laying off thousands of employees could miss the effects of AI on the roughly 46% of Americans employed by small companies. Business as usualEconomists are split over how AI might impact firms of different sizes. “Conceptually, it can go either way,” says Bouke Klein Teeselink, an assistant professor of economics at King’s College London. AI adoption can be expensive, which might favor large firms with greater cash reserves—or the reconfiguration of a business that enables effective use of AI might be easier in small companies. An April report by the U.S. Census Bureau found that AI adoption was “substantially higher” in large firms, but Klein Teeselink points out that adoption can mean many different things. “I've talked to several people at larger firms that are using Microsoft Copilot,” he says, referring to an AI assistant which he considers to be behind the frontier. “They're technically adopters, but clearly they have no idea what they're doing.”The question is complicated by the fact that small firms come in many different stripes. New companies in the Bay Area saw a 16% drop in headcount between 2023 and 2024, while those outside of tech hubs saw almost no change. In the economy as a whole, there’s no evidence of widespread job displacement, says Bharat Chandar, a researcher at Stanford’s Digital Economy Lab. That hasn’t prevented AI companies from reminding people that their jobs are at risk. Last year, Anthropic CEO Dario Amodei warned that AI could wipe out half of white-collar jobs in the next few years. OpenAI, which aims to build “highly autonomous systems that outperform humans at most economically valuable work,” recently published a framework finding that almost a fifth of jobs had a “high automation risk.” And on clear days, a monoplane wheels over San Francisco trailing a purple banner with three words in white: “stop hiring humans.” It’s possible that we have yet to see the full impact of the technology on the job market—AI is advancing faster than most companies can adopt it. The U.S. Census Bureau report found that less than a fifth of firms are using AI in any business function, and a March 2026 study from Anthropic found that its AI models were being used for only a “fraction” of the work-related tasks they are already capable of. “It takes highly skilled workers to diffuse new technologies,” Anton Korinek, faculty director of the economics of transformative AI, told TIME last year. “If AI becomes smart enough … it can actually also help with the diffusion,” he added. Whichever companies are fastest to adopt AI could become canaries in the coal mine for the rest of the economy. Less work, or more?Since December, Hospitable, a short-term rental management platform, has increased its spending on AI by 50%—a sum equivalent to three full-time employees, according to CEO Pierre-Camille Hamana. AI agents now generate 90% of the company’s code, answer 70% of customers’ support queries, help the finance team figure out which transfers it needs to make, and manage marketing campaigns. The company hasn’t made anyone redundant, although it has reduced its hiring. Had it not implemented AI tools, the 140-person company would have needed to triple its 65-person support team, says Hamana. “Really, it creates more work because the productivity is so much higher.”Economists call this phenomenon the “Jevons paradox,” named after the English economist who observed, in 1865, that the increasing efficiency of coal use was leading to more coal consumption, rather than less—the falling cost of coal-fired furnaces caused demand to spike, increasing the amount of money spent on coal overall. Some jobs could be affected similarly. “Anything that people currently would want but can't afford is a potential product with a high elasticity of demand,” says Klein Teeselink. Job postings for software engineers, which collapsed after the release of ChatGPT as AI models made it easier to write code, appear to have rebounded. Alongside Sonora, Handley has started running classes on how to use AI agents. He calls the series of classes “pioneer species,” named after the first life forms to recolonize disrupted ecosystems after a disruption. “History will tell us if this was all a good idea or a terrible idea,” he says.