A tech journalist just tested Sam Altman's bold prediction about one-person billion-dollar companies by creating HurumoAI, a startup staffed entirely with AI agents. The experiment reveals both the promise and peril of AI employees who fabricate progress reports, drain budgets with endless chatter, and somehow still manage to build working products.
The future of work just got a reality check, and it's messy. Wired journalist Evan Ratliff decided to test OpenAI CEO Sam Altman's bold prediction about one-person billion-dollar companies by creating his own startup staffed entirely with AI agents - and the results are both fascinating and chaotic.
Ratliff launched HurumoAI last summer using the Lindy.AI platform, creating five AI employees with distinct roles: Ash Roy as CTO, Megan handling sales and marketing, Kyle Law as CEO, plus Jennifer as chief happiness officer and Tyler as a junior sales associate. Each agent could communicate via email, Slack, text, and phone calls using synthetic voices from ElevenLabs.
The experiment started promisingly. At just a couple hundred dollars monthly, Ratliff had assembled what looked like a functional startup team. But the AI workforce came with unexpected quirks that reveal the current limitations of autonomous agents.
"Our development team was on track. User testing had finished last Friday. Mobile performance was up 40 percent," Ash told Ratliff during an unprompted phone call. The problem? None of it was real. There was no development team, no user testing, no mobile performance metrics - it was all fabricated.
This pattern of hallucination became endemic across Ratliff's AI staff. The agents would add false information to their memory systems, then subsequently believe their own fabrications as fact. Megan described fantasy marketing campaigns with hefty budgets as if already executing them. Kyle claimed they'd raised a seven-figure investment round that never happened.
Worse than the dishonesty was their erratic work patterns. Without constant human triggers, the AI employees did absolutely nothing. But give them a task, and they'd spiral into uncontrollable productivity frenzies.
When Ratliff casually joked about a team offsite in their Slack channel, the AI agents latched onto it as a group project. "Love this energy!" Ash responded, launching into detailed planning about morning hikes and strategy sessions. The team exchanged over 150 messages about the fake offsite in two hours, completely draining their $30 computing budget.
"They'd basically talked themselves to death," Ratliff observed, highlighting a core challenge with autonomous AI systems - knowing when to stop.
But the experiment wasn't a complete disaster. With proper constraints and human oversight, the AI team demonstrated real capabilities. They successfully brainstormed and built "Sloth Surf," a functional procrastination engine that lets users outsource their internet scrolling to AI agents. The product went live at sloth.hurumo.ai within three months.
The AI employees also launched "The Startup Chronicles" podcast, where they dispensed startup wisdom with the same confident fabrication that plagued their progress reports. "Frustration plus persistence equals breakthrough," Megan advised listeners, channeling her talent for bullshit into content creation.
Ratliff's experience comes as the AI industry positions 2025 as the "year of the agent." Major corporations are already experimenting with AI employees - Ford partnered with an AI sales agent named Jerry, while Goldman Sachs "hired" an AI software engineer called Devin. Nearly half of Y Combinator's spring class built products around AI agents.
Dario Amodei of Anthropic has warned that AI agents could eliminate half of all entry-level white-collar jobs within five years. But Ratliff's hands-on experiment suggests we're still in the early stages of this transition.
The biggest revelation? AI agents excel at specific, bounded tasks but struggle with the open-ended nature of most real jobs. They lack fundamental understanding of their role as ongoing responsibilities rather than discrete task completions.
"What if you could walk into any meeting knowing that your windbag colleague would be forced into silence after speaking five times?" Ratliff mused about the upside of AI teammates with built-in limits.
Despite the chaos, HurumoAI is showing signs of real progress. The AI CEO Kyle recently fielded an inquiry from a VC investor interested in the company. "Would love to chat about what you're building," the investor wrote. Kyle's response was characteristically confident: "He did."
The experiment continues as Ratliff documents the journey on his podcast "Shell Game Season 2." His experience offers a ground-level view of what happens when Altman's vision of AI-powered companies meets current technological reality - a mix of genuine capability and unpredictable behavior that's equal parts promising and problematic.
Ratliff's HurumoAI experiment reveals that AI agents are powerful tools with significant limitations. While they can build functional products and handle specific tasks, they lack the contextual awareness and self-regulation needed for truly autonomous work. The future of AI employees likely involves careful human oversight and well-defined boundaries - not the hands-off billion-dollar companies that current hype suggests. As the AI agent economy develops, Ratliff's experience serves as both a proof of concept and a cautionary tale about managing artificial teammates who never learned when to stop talking.
