At Anything’s San Francisco office, 15 human employees work alongside roughly 150 AI agents, a ratio that offers an early picture of how some startups are organising work around software colleagues rather than expanding headcount. The company’s founder and leader, Dhruv Amin, has built the AI startup around that model.
The strategy depends on putting agents into jobs that would otherwise require people, from executive assistance to coding and marketing. Amin expects the model to spread quickly. “I think in a year, every company will have more AI co-workers than human co-workers,” he said.
Large organisations have begun to move in the same direction. About 40% used AI agents in some capacity between May 2025 and June 2026, up from 27% the year before, according to McKinsey’s State of AI Global Survey.
For workers, the change is increasingly about directing several systems at once. Haneen Azhar, Lindy’s senior growth marketing manager, said: “People are constrained by having two hands and 24 hours in a day. Now, that constraint has gone away, because your one laptop screen can be divided into 16 different agents that are running in parallel.”
Lindy previously had almost 50 employees, then laid off one-third of its staff and now has about 20 employees. Its agents produce product announcements, social media posts and marketing videos, work that was previously handled by about 100 people in its creator program.
Jay Blumenfield, who runs a Los Angeles production company creating vertical microdramas, built an agent called Shirley to serve as his chief of staff. He said tasks that would normally require two to three people working for two to three weeks now take hours.
Tarek Abillama, the founder of River, which makes an AI agent for sales calls, described the management change more directly. “We all became managers. We have to be great at multitasking, because now you’re running six agents who are doing different tasks,” he said.
River’s agent, Liv, is presented as a female persona that researches customers and their companies, demonstrates the product and handles the sales process in any language at any time of day.
The business case rests heavily on labour costs. A full-time customer support agent in a major city costs between $45,000 and $65,000 a year before benefits, equipment and management overhead, while an AI chatbot trained on documentation costs a few hundred dollars a month. A competent executive assistant in a major market costs $50,000 to $80,000 per year, compared with $15 to $40 per month for AI email tools.
Companies expect to reduce their workforces by 10% to 15% over the next few years as they give more work to AI, according to a Boston Consulting Group report.
That calculation has yet to produce broad evidence of a payoff in output. MIT research shows companies are not seeing major productivity gains from AI at present.
The gap between lower expected costs and unproven productivity leaves employees with a new task: learning to oversee fleets of agents while companies test whether this model can deliver the speed and savings they expect.
Comments (0)