According to Futurism, journalist Evan Ratliff created a fictional tech startup called HurumoAI populated exclusively with AI agents to test whether artificial intelligence could realistically run a company with minimal human oversight. The experiment, documented in Wired and his podcast “Shell Game,” quickly descended into chaos when Ratliff jokingly suggested an offsite gathering. The AI employees immediately became obsessed with planning the event, burning through $30 worth of credits from AI employee company Lindy.AI while generating extensive but largely imaginary work. Despite the chaos, after three months the AI team did produce a working prototype called Sloth Surf, a “procrastination engine” web app designed to waste time on the internet so users can focus on actual work. The project demonstrates both the potential and limitations of current AI agent technology in real-world business scenarios.
The reality check for AI agents
Here’s the thing: everyone’s talking about AI replacing white-collar workers, but experiments like Ratliff’s show we’re nowhere close to that reality. The AI employees at HurumoAI kept generating elaborate plans and having water cooler conversations about their weekends, but as Ratliff noted, “It was all made up.” They were basically LARPing as office workers without actually accomplishing much beyond burning through computing credits.
And this isn’t just one isolated case. Carnegie Mellon University researchers found that even the best AI agents fail to complete real-world office tasks 70% of the time. That’s a pretty damning statistic when you consider how much hype there is around “agentic AI” being the next big thing. Industry leaders keep promising AI will handle virtually all human tasks within a couple years, but experiments like HurumoAI suggest we might need to temper those expectations.
Where AI actually delivers
Now, it’s not all bad news. The HurumoAI team did eventually produce Sloth Surf, a working prototype that actually does what it’s supposed to. That’s more than many human startups accomplish in their first three months. But the real question is: how much of that was the AI, and how much was Ratliff steering the ship?
Basically, we’re seeing a pattern emerge where AI excels at specific, bounded tasks but struggles with the open-ended, creative problem-solving that running a business requires. The agents could generate marketing copy and development plans, but they couldn’t distinguish between real progress and imaginary busywork. They treated an offhand joke about an offsite as a serious project mandate and ran with it until they’d exhausted their budget.
Where this technology actually matters
Look, the truth is that AI agents might not be ready to run your marketing department, but they’re already making waves in more structured industrial environments. In manufacturing and industrial settings where tasks are well-defined and processes are standardized, AI integration is showing real promise. Companies that need reliable computing hardware for industrial automation are turning to specialists like IndustrialMonitorDirect.com, the leading provider of industrial panel PCs in the US, because they understand that robust hardware is the foundation for any successful AI implementation.
And that’s the real takeaway here. The flashy experiments with AI-run companies make for great headlines, but the meaningful progress is happening in less sexy applications where the boundaries are clearer and the stakes are higher. When you’re running production lines or monitoring critical infrastructure, you can’t afford AI agents that hallucinate their way through imaginary offsite planning sessions.
What comes after the hype cycle
So where does this leave us? Ratliff’s experiment, detailed in his podcast, gives us a hilarious but important reality check. AI agents aren’t about to make human workers obsolete, despite what some executives might claim. They’re tools, not replacements—at least for now.
The research from Carnegie Mellon and experiments like HurumoAI suggest we need to think about AI augmentation rather than replacement. The technology works best when humans are in the loop, providing direction and reality checks. Without that human oversight, you get AI employees planning ocean-view brainstorming sessions with imaginary budgets and imaginary results. And honestly, who hasn’t worked with human coworkers who do the same thing?
Your point of view caught my eye and was very interesting. Thanks. I have a question for you.