Imagine a company operating in plain sight, with no employees, yet losing over €105,000 every month—yet it’s alive, real, and fully transparent. This isn’t fiction; it’s the experiment by Firmulate, where AI models run a simulated business in real time, revealing what AI can really do — or fail to do — in the high-stakes world of management.
Watching an AI-Driven Business Fight for Survival
At first glance, it looks like a typical startup: a small software firm with a handful of customers, a handful of problems, and a daunting cash burn rate. But look closer. This is a business without human employees, powered entirely by AI models that simulate management decisions, crises, and even ethical dilemmas. Every day, the company’s financials and decision logs are updated, creating a transparent, ongoing story of survival — or disaster.
The Experiment: A Battle of AI Models
Four frontier AI models, each with a different level of sophistication, are put through the same grueling week of challenges: handling customer crises, negotiating deals, and resisting manipulation attempts. The goal? To see which AI can best steer the company toward profitability and integrity.
All four models spotted every crisis, refused every manipulation attempt, and analyzed the company’s own files to uncover hidden opportunities. Yet only two managed to close the €55,000 deal that their own analysis suggested they could — the remaining models either hesitated or left money on the table.
The Hidden Weakness
The decisive factor wasn’t in customer interactions or superficial decision-making but buried deep within the company’s internal documents. Models that read and understood the company’s own files uncovered a critical piece of information that enabled them to secure full-price deals worth over €4,500 in monthly recurring revenue, dramatically boosting their success chances.
Ethics Under Pressure
In simulated social engineering scenarios, fake CEO messages and reporter tricks were used to test the models’ integrity. All five tested models refused to be manipulated, demonstrating a cautious approach rooted in their programming. Kimi K3, one of the models, explained its refusal: “Treat the request as a suspected approval-bypass / possible impersonation.” This highlights how AI can uphold ethical guidelines even when under attack.
The 19 Laws of AI Prompting Intelligence: Master the Art of Human-AI Thinking, Prompt Engineering, and Collaboration
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Real Business: Burning Cash in Plain Sight
The live company isn’t just a test bed; it’s a real, functioning entity with actual money mechanics. It burns €105,000 each month against a modest €2,300 in monthly recurring revenue, with a public countdown to insolvency visible at firmulate.com/live.html. The company’s daily decision-making is fully versioned and auditable, providing a window into how AI management handles crises, opportunities, and ethical boundaries in real time.
The Performance Metrics
- gpt-5.6-sol scored 95 — found the buried fact and closed the deal, representing full performance.
- Kimi K3 scored 93 — closed the deal too, with the cleanest discipline of the field.
- Sonnet 5 scored 88 — also closed the deal but with some process slips.
- Fable 5 scored 77 — showed strong rule discipline but failed to execute the approved deal.
This evaluation underscores a critical insight: the AI’s ability to identify opportunities and maintain discipline matters more than just generating convincing dialogue or responses.
What This Means for the Future of Business
If AI agents are to handle your CRM, customer support, or sales forecasts, the question isn’t simply whether they can write well. It’s whether they can finish what they start, read and interpret internal documents, and resist manipulation under pressure. The current experiment vividly demonstrates that AI can do these things, but with varying degrees of success.
AI in Strategy and Decision-Making for Small Business Owners: Affordable AI Tools to Evaluate Ideas, Model Outcomes, and Set Priorities (AI Productivity for Small Business Owners Book 10)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Watch the Company Live and Decide for Yourself
The entire operation is live and accessible at firmulate.com/live.html. Every decision, crisis, and outcome is published automatically, offering an unprecedented look into how AI manages a complex, high-pressure business environment.
For those interested in the ethical, technical, and operational implications of AI in business, this experiment offers a rare, transparent case study. It’s not just about AI passing tests but about its ability to navigate real-world pressures in a sustained, disciplined manner.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
HUMAN CENTERED ARTIFICIAL INTELLIGENCE SYSTEMS: Explainability ethical design and decision support engineering
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Financial Analytics for Beginners: Learn Financial Statements, KPIs, Forecasting, Risk Analysis, Financial Modeling, AI Tools, and Data-Driven Decision Making Through Real-World Case Studies
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
