Some startups use AI to cut head counts or contractors

LinkedIn: 2022 startups averaged 20.7 workers by year three

A small group of early-stage founders has begun asking, before each new hire, whether AI can do the job instead, The Wall Street Journal reported. Some have answered by laying off employees or ending relationships with contractors and agencies. The practice extends a years-long trend toward leaner startups: average head counts at U.S. startups have fallen from a 2021 peak, and startups founded in 2022 with at least five employees in their first year averaged 20.7 workers by their third year, according to LinkedIn analysis cited by the Journal. By comparison, similarly sized startups founded in 2018 had an average of 32.5 employees in year three.

Plenty of startups still increase head count at a rapid clip as they aim for billion-dollar valuations and fight for top engineering talent. But the subset of founders leaning hardest on AI is embracing a different ethos: stay small for as long as possible.

Simar Singh, co-founder of Butternut AI, which builds a no-code website builder, told the Journal that he and his co-founder shed five engineers at the end of last year after determining AI agents could do the work around the clock. Butternut AI is now down to four employees from a peak of nine and continues to grow, Singh said, with around 10,000 paying customers. “We are four employees,” Singh said. “We were nine at our peak.”

Lindy, a startup with around two dozen employees that has raised $50 million, makes an AI “teammate” that integrates with customers’ Slack workspaces. The company laid off a handful of marketing employees in January as it pivoted its product and never refilled the roles, Haneen Azhar, a senior growth marketing manager at Lindy, told the Journal. Even as Lindy’s user base grew to half a million, the marketing team held at two people, supplemented by AI-agent helpers. Lindy is still hiring coders. Work that once required full-time data engineers now goes to prompts, Azhar said. Lindy AI used to pay five outside creative agencies for content and video work; that count is now zero, with AI video generation taking their place. “We’ve been able to cover all the surface area that we used to,” Azhar said. “AI left and right.”

Startup accelerator Y Combinator says the lean approach can yield success, the Journal reported. Emergent, which went through Y Combinator’s program in 2024, reached $15 million in annualized revenue with 15 people. Retell AI, which went through the same year, hit $60 million in annualized revenue with a head count of around 40.

AgentCollect, which uses AI agents for invoice collection, has shrunk from a peak of 50 employees to about 30, the Journal reported. Roles largely automated include billing coordinators, quality evaluators for customer calls and a product manager. Chief Executive John Banner framed the approach as a hiring philosophy, not a cost-saving move: employees should do work that requires creativity and judgment. “This is the thing that as of today cannot be delegated to AI,” Banner said.

The lean approach has limits, according to venture investors interviewed by the Journal. Liz Wessel, a partner at First Round Capital, said whether a startup can stay tiny depends on what it sells and how fast revenue grows; companies that are hyperscaling still compete for top engineering talent. Aileen Lee, founder of Cowboy Ventures, said two co-founders can now build prototypes, deliver to customers and test what they’re building before hiring — but only up to a point. “If you are selling something that’s $60,000 or $75,000, and selling it to midmarket or enterprise customers, they want to be buying from a company that has humans,” Lee said.

Investors themselves can pressure lean startups to add staff. To raise a Series A, Lee said, venture capitalists generally expect $4 million to $5 million in revenue in the first year — a target that is “really hard to achieve with no people.”

Some founders have taken the lean approach further still. Advait Paliwal, who has not yet been meeting with venture capitalists, is operating solo, building a consumer mobile app that lets people talk to a variety of AI agents; he considers OpenAI’s Codex his chief technology officer. Paliwal runs 20 Claude and Codex subscriptions rather than hiring employees, and over the past year he interviewed four software engineering candidates — including a 14-year-old in Belgium — without making an offer. “The amount of instruction I had to give them was the same amount of instruction that I could just give Claude,” Paliwal said. “And it would just answer faster.”