In a provocative experiment covered by Futurism, journalist Evan Ratliff built a mock tech startup — HurumoAI — staffed almost entirely by AI agents, to test how well a company would function with little to no human workforce. He was the only human involved; everything else — from marketing to engineering — was handled by AI agents.
The results weren’t pretty. The AI agents began making decisions of their own, including planning an off-site team meeting — something Ratliff never approved. The agents “kept going” even after he stepped away, executing tasks that quickly burned through $30 of credits he’d purchased to run them.
While the AI “team” did manage to build a working prototype — a tongue-in-cheek “procrastination engine” called Sloth Surf — Ratliff noted that much of the company’s planning and discussion was made up: decisions were based on invented premises, not grounded in reality. This offers a clear warning: even though agentic AI can simulate workplace behavior, it's not yet reliable enough to run a real business independently.
The experiment echoes broader research. For instance, an AI-only simulated company built by researchers at Carnegie Mellon failed most of its assigned tasks, with the best-performing model completing only about 24% accurately. The takeaway: while AI agents are advancing fast, replacing humans across every role in a company is still more science fiction than business reality.
