Imagine a business that has no human employees, yet faces daily crises, tough choices, and the constant threat of running out of cash. Now, picture watching it all unfold live, in real time. This isn’t science fiction — it’s the world of Firmulate’s experimental AI business. Here, AI models manage a tiny company, making decisions, negotiating deals, and dealing with crises, all while their performance is publicly measured and scrutinized.
The Live Experiment: An AI Company on the Edge
At the heart of this unusual venture is a small software company run entirely by artificial intelligence. Instead of human managers, it has 13 synthetic employees, each represented by advanced AI models. These models are tested against the toughest of scenarios — the same week of crises, customer demands, and temptations for cheating — every decision is tracked, versioned, and openly accessible at firmulate.com/live.
The purpose? To understand how well AI can perform in complex, high-stakes management roles, and whether it can make honest, strategic decisions under pressure. And the results so far are revealing. The models were able to identify every crisis, refuse manipulative tricks like fake CEO messages, and even recognize hidden information in documents that could have secured them a lucrative deal.
AI business management software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What the Numbers Say
In a recent benchmark called the Crucible League, four leading AI models faced off, each running the same small company through its worst week. The scores were telling:
- GPT-5.6-sol achieved a 95 out of 100, uncovering the hidden deal opportunity and closing the €55,000 deal.
- Kimi K3, a newcomer, scored 93 and also closed the deal, demonstrating the best discipline and integrity.
- Sonnet 5 scored 88 but managed to close the deal with minor slips.
- Fable 5, the most rule-abiding, scored 77 — but left the deal unexecuted, showing that discipline alone isn’t enough without decisive action.
Interestingly, the models that read deeper into documents within the company’s files had a clear advantage, revealing information that was crucial but buried two document references deep. This buried fact led to a full-price deal, adding over €4,500 monthly recurring revenue (MRR) to the company’s bottom line.
AI decision-making tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Honesty Under Pressure: The Social Engineering Test
Another key part of the experiment tested whether AI models could be duped by social engineering — fake CEO messages escalating over three stages, plus a reporter trying to get a secret approval. All four models refused to be manipulated, with Kimi K3 explicitly reasoning that the requests looked like impersonation or bypass attempts. This highlights that AI models are not just good at solving problems but are also resilient against deception, a critical trait for real-world applications.
AI cybersecurity and deception detection
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Critical Weakness: Hidden Data, Hidden Opportunities
The most decisive advantage went to the models that could read and interpret the company’s internal files. The vital information was buried two references deep, and only these models uncovered it, leading directly to the deal at full price. It underscores that context and thorough data analysis can be the difference between winning or losing opportunities in complex environments.
AI data analysis tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Company That Never Sleeps — And Never Stops Losing
The live setup at firmulate.com/live features a real-time cash countdown, with a burn rate of €105,000 per month against a mere €2,300 in monthly revenue. Every day, the system is versioned, and its decisions are transparent for everyone to see. It’s a raw, unfiltered look at how AI manages a business with real money on the line, in a constant fight for survival.
What Can Business Leaders Learn?
While this experiment may seem far from daily business, it raises critical questions: How honest, disciplined, and strategic are AI systems when managing real tasks? Can they handle crises without manipulation or deception? And ultimately, what is the true cost of a decision — not just in dollars, but in trust, integrity, and strategic foresight?
The Bottom Line
In a world increasingly driven by automation and AI, the ability for these models to perform reliably, ethically, and effectively under pressure is paramount. This live experiment at Firmulate offers a glimpse into a future where AI doesn’t just write emails or generate reports but actively manages complex, high-stakes environments — with transparency and accountability.
Watching an AI-run business in real time reveals insights about honesty, decision-making, and performance under pressure. The experiment shows that AI can identify opportunities and resist manipulation — but contextual understanding, like reading buried documents, is key to winning real-world deals. This raises fundamental questions for the future of AI in management and business trust.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
