
In a world where emergency preparedness often depends on human decision-making and manual processes, what if you could test your company’s resilience against crises using artificial intelligence? Imagine a company that has no employees, burns through €105,000 each month, yet is publicly fighting for survival, all while you watch every decision unfold in real time. This is not science fiction—it’s the live experiment from Firmulate.
The Live Company That Runs on AI and Public Scrutiny
At the heart of this experiment is a small, synthetic company operated entirely by AI models. It’s a real-time showcase where 13 simulated employees handle crises, make decisions, and try to sustain a business with a monthly recurring revenue (MRR) of just €2,300, while burning through €105,000 in operating costs each month. This stark economic imbalance underscores the company’s fragile survival, making it a compelling case study for risk management and operational resilience.
How the AI Company Works
This virtual business is constantly monitored, with every decision versioned and auditable. It faces the same crises, customer demands, and temptations that a real company would encounter—only this time, the responses come from AI models that have learned over 680+ self-created rules from a public playbook. The experiment is updated twice daily, and anyone can watch the company’s daily struggles unfold at firmulate.com/live.
The Disruptive Findings
What makes this experiment intriguing is its transparency and the findings from testing different AI models—each representing a different decision-making style—against the same set of challenges. The most advanced model, GPT-5.6-SOL, scored a remarkable 95 out of 100, successfully identifying hidden information in documents that led to closing a €55,000 deal, which would boost the company’s MRR by over €4,500. Meanwhile, other models, like Kimi K3 and Sonnet 5, scored slightly lower but still managed to close deals, demonstrating solid discipline and decision-making under pressure.
The Challenge of Manipulation and Trust
In simulated social engineering attacks—such as fake CEO messages and reporter tricks—all models refused to be manipulated, illustrating their capacity for integrity. Kimi K3, for example, explicitly treated suspicious requests as potential impersonation attempts, showing a level of on-record reasoning that’s vital for real-world applications. Yet, despite these capabilities, none of the models could fully replicate the human nuance needed to execute some deals or escalate issues consistently, highlighting ongoing limitations.
AI decision-making simulation software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Bigger Picture: Resilience, Trust, and Cost
This experiment is about more than AI decision-making—it’s a raw look at what it takes to run a business under extreme conditions while maintaining honesty and discipline. For survival-minded organizations, especially those in emergency or crisis prep scenarios, the question isn’t whether AI writes well, but whether it can see through crises, avoid shortcuts, and stay committed to long-term goals.
Lessons for Emergency Preparedness
Just as survivalists prepare for unpredictable threats, companies exploring AI decision-making must understand how these systems behave under pressure. The Firmulate experiment shows that, with the right models, AI can detect hidden vulnerabilities, refuse manipulation, and make disciplined decisions—even when stakes are high and resources are limited.
Real-World Application
For enterprises interested in stress-testing their own operations, Firmulate offers a ‘wargame’ platform that simulates business crises without risking actual systems. These tests can reveal decision weaknesses, misaligned priorities, or vulnerabilities to manipulation—insights crucial for building resilience in uncertain environments.
business resilience stress test tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Road Ahead
While the current models demonstrate promising discipline and crisis detection, the experiment also exposes gaps—such as missed opportunities or incomplete execution—highlighting that AI decision-making still has room to grow. The live data, accessible at firmulate.com/live, serves as a real-world laboratory for understanding how AI can both bolster and challenge organizational resilience.

Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
AI cybersecurity social engineering defense
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
enterprise AI risk management tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.