AI agents as co-workers: inside startups betting on bot-staffed offices

At the San Francisco AI startup Anything, Dhruv Amin gets through his workday by messaging colleagues: Gwen schedules his meetings, Cody helps troubleshoot technical problems, Betsy handles search engine optimisation and Juno looks after customer onboarding. When the office runs short of snacks, he asks Justin, the office manager, to restock the kitchen.

None of them are people. They are AI agents – software that carries out tasks without human intervention. Anything, which has 15 human employees and roughly 150 agents, is one of several startups betting that companies will hand more sales, marketing, coding and other work to semi-autonomous software.

"I think in a year, every company will have more AI co-workers than human co-workers," Amin told The Guardian.

A status symbol, and a source of tension

In San Francisco's tech scene, agents have become a status symbol, showing that a company can do more with less, according to several startup employees. Workers routinely boast about how well their agents handle emails or sales pitches. But the technology also causes friction: it alienates customers and colleagues who want human contact, removes roles, and turns employees into overnight managers of bots that often make mistakes.

Corporate America may not be far behind. About 40% of large organisations used AI agents in some capacity between May 2025 and June 2026, up from 27% a year earlier, according to McKinsey's State of AI Global Survey. Boston Consulting Group found companies expect to shrink their workforces by 10% to 15% over the next few years as they delegate more to AI, while MIT research shows firms are not yet seeing major productivity gains. Vendors including Viktor, Lindy, OpenAI's Dots, Anthropic's Claude Cowork and Microsoft's Scout are selling the idea of automated co-workers.

'We all became managers'

Lindy, which once employed nearly 50 people, laid off a third of its staff and now has about 20, filling many open roles with agents. Haneen Azhar, its senior growth marketing manager, has offloaded calendar management, marketing tests and brand images to agents, and says they now handle product announcements, social posts and videos that about 100 people in its creator programme once did. She worries about jobs and about new entrants to the workforce: "If I was trying to get my foot in the door right now, I'd be worse off than I was a year ago."

In Los Angeles, producer Jay Blumenfield replaced a fired assistant director on a microdrama shoot with an agent that produced the schedule, call sheet and day's script and sent them to cast and crew. "It would have been an unmitigated disaster, but this kind of saved us," he said. He and his partner, Tony Marsh, now run the company alone. "The game in Hollywood is that you're not allowed to talk about using AI," he said. "I don't want it to replace people. But we should be honest about how it is helping us."

Agents still need supervision. Rafael López Muñoz of the startup Wordware said he spent most of his time undoing errors after letting agents run recruiting, including mix-ups between candidates with similar names and hallucinated résumé details. Tarek Abillama, founder of River, said: "We all became managers." Its sales agent, Liv, receives flirtatious messages and off-script requests for discounts, he said.

Brooke Hopkins, founder of the AI voice startup Coval, has banned agents from customer communication, arguing it is lazy and sends the wrong message. Duncan McClure, a contractor who evaluates AI models, said agents hounded him with recruiting emails and calls: "The message it sends is that we don't care enough to actually put in any effort," he said, adding, "it's my time that they're wasting."

Amin says he gets more done with agents – but he has not stopped working weekends.

Source: The Guardian.