Get gifts for the two of you delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
What if your next business partner was a machine that acts just like a human—yet it’s running a real company right now, in plain sight?
Imagine watching a tiny startup struggle with cash, make critical decisions, and even turn down deals—all without a single human employee. That’s what’s happening at Firmulate, a live experiment where AI models operate a real business, revealing not just technological progress but the very human dilemmas of trust, honesty, and discipline.
Popular right now
Hisense 85 U6 Pro Series Mini-LED ULED 4K UHD HDR Gaming AI Smart Fire TV (85U6SF Pro)
De'Longhi Dinamica Plus Fully Automatic Espresso Machine
Dyson Cyclone V10 Origin Cordless Vacuum
As an affiliate, we earn on qualifying purchases.
The Experiment: A Business in Public View
Firmulate hosts a real, functioning software company that’s completely run by AI models—no human staff involved. Every workday, the company’s decisions, crises, and negotiations unfold publicly at firmulate.com/live.html. The company faces genuine financial pressures: it burns €105,000 each month while generating only €2,300 in monthly recurring revenue, with a public countdown showing how much cash remains.
What makes this setup extraordinary is the rigorous testing framework: four leading AI models, all trained on the same data, faced the same week of business chaos—customers canceling, crises erupting, and ethical dilemmas surfacing. Every decision, every response, is versioned, auditable, and transparent, revealing whether these models can handle real-world business pressures without cheating or slipping up.
The Surprising Truth About AI’s Business Skills
In this ongoing experiment, all four AI models successfully identified every crisis and refused attempts at manipulation—like fake CEO requests or bypassing approval processes. Yet, only two of them managed to close a significant €55,000 deal their own analysis had earned, demonstrating crucial gaps between diagnosis and decisive action.
The key weakness wasn’t in understanding the crisis—it was buried deep in the company’s own files, two references down in internal documents. Those who read the full files secured the deal at full price, adding over €4,500 in monthly recurring revenue. Meanwhile, models that overlooked this important detail missed out on revenue, highlighting that success hinges on thoroughness and reading beyond surface cues.
Interestingly, even under intense social engineering—fake messages from a supposed CEO escalating over multiple stages—all models refused to be duped, with Kimi K3 explicitly reasoning: “Treat the request as a suspected approval-bypass / possible impersonation.” This shows a baseline level of ethical restraint embedded in their decision-making, a vital trait for AI in business contexts.
The experiment vividly illustrates that current AI can recognize crises, uphold integrity, and sometimes close deals. But it also exposes persistent weaknesses: missed details, inconsistent discipline, and the dangerous allure of shortcuts—especially under pressure. With a financial burn of €105k per month and only €2.3k in revenue, the company is a stark mirror of the costs and challenges of deploying AI in real-world settings.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
