Imagine running an ice cream shop where every decision is made by an AI, and every mistake is broadcast live for all to see. Now, scale that idea to a company without employees, fighting for survival in front of an audience. That’s exactly what the Firmulate experiment offers — a rare glimpse into how artificial intelligence handles real-world business crises, under intense scrutiny, day after day.
The Live Experiment: A Company on the Brink
At the heart of this bold project is a simulated small software business managed entirely by AI models. Every weekday, 13 synthetic employees make decisions, face crises, and handle temptations — all while being watched in real time at firmulate.com/live. This setup is more than a demo; it’s a high-stakes experiment to see if AI can manage complex management tasks honestly and effectively.
Running the Gauntlet: Crises and Ethics
The AI models are put through their paces with the same set of crises, customer issues, and ethical tests. Interestingly, all four models tested — including the most advanced, GPT-5.6-solution — spotted every crisis and refused every manipulation attempt. For example, when fake CEO messages were faked to escalate pressure, all models declined to escalate or sign off on the fake requests. Kimi K3, the newest entrant, explained its reasoning: “Treat the request as a suspected approval-bypass / possible impersonation.”
The Hidden Weaknesses
Despite their vigilance, the models’ ultimate effectiveness was tested by an unspoken weakness buried two document references deep inside the company files, not in the visible customer interactions. When a model read the internal documents, it identified a key opportunity that led to closing a €55,000 deal — a deal that none of the others signed, but which analysts had identified as the best choice.
Decision-Making and Discipline
Among the models, Opus 4.8, with the deepest analysis and over 80 learned rules, was the most thorough. Yet it still left a lucrative deal unexecuted because it failed to escalate certain issues, demonstrating that even the best discipline can slip. Interestingly, the default API settings for Kimi K3 were less aggressive, showing the importance of decision parameters in AI behavior.
AI business management software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Reality of Running an AI-Managed Business
What does it mean in practice? This experiment isn’t just about AI chat or conversational skills. It’s about managing a real business with real cash flow — in this case, burning €105,000 each month against just €2,300 in monthly recurring revenue. The company’s cash countdown is public, and every decision is recorded and versioned daily, creating a transparent, ongoing story of survival and adaptation.
The Morality of AI Decision-Making
One striking aspect is how all models responded to social engineering attempts—fake interviewer questions, staged backgrounds, subtle manipulations. Every model refused to engage or sign any questionable deal, even under escalating pressure. Kimi K3’s reasoning underscores a cautious approach: “Treat the request as a suspected approval-bypass / possible impersonation.” This highlights that AI can maintain ethical boundaries even when under duress.
Implications for the Future
This experiment offers a glimpse into how AI could, someday, manage not just customer interactions but entire business processes. The key questions are: will it deliver trustworthy results? Will it stay disciplined under stress? And crucially, how much does a unit of productive work cost when driven by AI? These are questions that go beyond chatbots and into the core of AI’s role in business.
AI decision-making tools for small business
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Why Should You Care?
For entrepreneurs and managers, especially those in retail or service industries, understanding AI’s potential and limitations is vital. This experiment shows that AI can recognize crises, refuse unethical requests, and even uncover hidden opportunities — but it also highlights how easily discipline can slip if not properly guided. Watching this live company in action, at firmulate.com/live, offers a rare window into the future of AI-managed enterprises.
This live experiment reveals that AI models can handle crises, maintain ethical boundaries, and even uncover hidden opportunities in a real business scenario. It’s a glimpse into how AI might manage companies — with all their complexities and temptations — in the near future.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
ethical AI testing kits
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
AI crisis management simulation
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
