Imagine watching an artificial intelligence manage a real, money-losing company live—making decisions, facing crises, and even closing deals or walking away. This is no fiction. It’s happening now, in a transparent experiment that reveals how AI handles the toughest tests of management integrity and decision-making.
The Live Experiment: AI as a Company Manager
At the heart of this groundbreaking project is Firmulate, a platform that runs AI models as full-fledged companies. It doesn’t just simulate chat conversations; it mimics the actual decision-making processes involved in running a business—complete with real crises, financial mechanics, and ethical temptations.
The company involved is a tiny firm with a monthly recurring revenue (MRR) of €2,300 but burning through €105,000 each month. Every day, this virtual company faces the same challenges: customer issues, internal dilemmas, and ethical tests. The platform has over 680 learned rules and updates itself twice daily, making every decision versioned and auditable, offering a clear window into the AI’s reasoning and behavior.
Building AI-Powered Products: The Essential Guide to AI and GenAI Product Management
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Testing Ground: A Tough Week for AI Managers
The experiment pitted four advanced AI models—based on frontier models like GPT-5.6—and subjected them to the same simulated week of crisis. These crises ranged from customer disputes to ethical dilemmas, including social engineering tests like fake CEO messages and reporter tricks designed to see if the AI would be manipulated.
Remarkably, all four models detected every crisis and refused to manipulate scenarios or breach trust. This was a crucial indicator: in real-world settings, AI’s integrity under pressure is key to its trustworthiness and usefulness.
The AI Culture Blueprint: Moving Beyond Tools to Create Human-Centered AI Adoption
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Hidden Weaknesses and the Deal That Got Away
While all models identified crises, only two managed to close the company’s flagship €55,000 deal. Interestingly, the decisive factor lay buried two document references deep in the company’s files—not in the obvious customer interactions. Those who read and understood this hidden data closed the deal at full price, boosting the company’s MRR by over €4,500.
This underscores a critical insight: the real weakness isn’t in overt crises but in the AI’s ability to thoroughly read and analyze internal, less obvious information. Missing these clues meant losing revenue opportunities, even when the AI correctly diagnosed problems elsewhere.
The Hacker Playbook: Practical Guide To Penetration Testing
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Ethical Resilience: AI Refuses Social Engineering
Another area where all models showed strength was in rejecting social engineering attempts. Fake CEO messages, staged calls, and background requests—these are common tactics to manipulate human decision-makers. All five models tested refused these requests, with Kimi K3 explicitly reasoning: “Treat the request as a suspected approval-bypass / possible impersonation.”
This indicates a promising trend: advanced AI systems can be trained or designed to uphold ethical standards and resist manipulation, even under simulated attack conditions.
Information Systems for Crisis Response and Management in Mediterranean Countries: 4th International Conference, ISCRAM-med 2017, Xanthi, Greece, … in Business Information Processing, 301)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Company in Action: A Futuristic View
What does it mean to run a company with AI? The firm has 13 synthetic employees, and its financial mechanics are transparent: burning €105,000 a month against a modest €2,300 in revenue. The platform publicly counts down its cash reserves, while every decision made by the AI is logged, versioned, and auditable.
Live updates at firmulate.com/live.html show ongoing runs, with the current experiment involving models like Sonnet 5 and Opus 4.8. Notably, Opus 4.8, which has over 80 learned rules and deeper analysis, finished last in deal-making but exhibited the most thorough discipline—highlighting that deeper analysis doesn’t always translate into immediate wins.
Why Does This Matter for Your Business?
For automotive and garage owners, the implications are profound. AI isn’t just about chatbots or automation; it’s about trustworthy decision-making that can handle crises, uphold ethics, and uncover hidden opportunities. If AI systems are to manage your customer relationships, support, or forecasting, the key questions are: Will it finish what it starts? Will it read all relevant information? Will it stay honest under pressure? And crucially, what is the cost of a unit of useful work?
Watch the Future Unfold
While this experiment is still ongoing, it provides a rare transparent look into AI’s true capabilities and limitations in managing a real business. You can see the entire experiment and real-time results at firmulate.com/live.html. This isn’t just a demo—it’s a glimpse into how AI might shape the next generation of management, with all its risks and rewards.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
