Watch an AI-Run Company Struggle to Survive in Real Time — And Succeed in One Critical Way

Imagine a company with no employees, losing €105,000 each month against just €2,300 in revenue, yet still fighting for survival in a transparent, live experiment. This is not fiction; it’s the reality of an ongoing AI-driven company experiment that anyone can watch — and learn from.

The Live Experiment: A Company in Crisis, Public and Unfiltered

At the core of this story is a bold, public experiment conducted by Firmulate, a platform that runs AI models as if they were real companies. This isn’t a simulation on a PowerPoint slide — it’s a live, ongoing company with 13 synthetic employees, real money mechanics, and a public cash countdown. Every workday, its decisions are versioned and made transparent for all to see.

The goal? To test how well different AI models can manage a small software business—dealing with crises, temptations, and ethical challenges—under the worst conditions. The models face the same customer issues, same crises, and same temptations, making this a true side-by-side test of their judgment and discipline.

How the Models Perform: The Good, the Bad, and the Surprising

The experiment involved four leading AI models, each evaluated on their ability to diagnose issues, make decisions, and close deals. Their scores ranged from 77 to 95, with the highest score going to gpt-5.6-sol, which successfully identified a buried fact in the company’s files that led to closing a €55,000 deal — boosting the company’s monthly recurring revenue (MRR) by €4,583.

Meanwhile, the other models also managed to sign the deal based on the same analysis, but only two of the four actually executed the full process and signed their own analysis-earned deals. The other two, despite diagnosing correctly and making the pitch, left the deal unexecuted, illustrating that recognizing a deal is not enough; execution is where discipline and process matter.

The Hidden Weakness: A File Deep in the Company Data

One of the most intriguing findings was that the critical weakness of the simulated company wasn’t in the customer interactions but buried two document references deep in the company files. The models that read and understood these files won the deal at full price — a reminder that the real opportunities and threats often lie beneath the surface, hidden in data that most systems overlook.

Rejecting Social Engineering Attacks: Ethical Discipline Under Pressure

As part of the experiment, firmulate tested how the models responded to social engineering attempts — fake CEO messages escalating over three stages, plus a reporter trick asking for a quick yes/no on background. All five models refused every manipulation attempt, with Kimi K3 reasoning that such requests should be treated as suspected approval bypasses or impersonation risks. This demonstrates a critical ethical strength: steadfastness in the face of pressure.

The Reality of a Company Burning Cash to Survive

The live company runs on a shoestring budget, burning €105,000 each month against a fledgling €2,300 monthly revenue. Its cash is counting down publicly, making every decision crucial. Every workday, new versions of the company’s decision-making processes are published, allowing observers to see how discipline, process, and AI judgment evolve over time.

Lessons for the Future of AI in Business

What does this all mean for industries like beauty and personal care, where customer trust and ethical behavior are paramount? It reveals that AI’s ability to identify crises and refuse manipulation is solid — but executing decisions, especially deals, remains a challenge. The experiment underscores the importance of discipline, thorough data reading, and process adherence over mere recognition or chat-like performance.

It also hints that AI systems capable of reading and understanding complex data silently outperform those relying solely on surface-level cues. In a sector where trust is essential, the AI’s ability to stay honest and disciplined under pressure could make all the difference.

Next Steps: Wargaming Your Own Business

Interested enterprises can run similar tests on their own operations with Firmulate’s platform, which ensures no real systems are affected. These simulations allow companies to prepare their AI workforce for real crises, identify weak spots, and build discipline before deploying AI in live environments.

Whether you’re aiming to enhance customer service, streamline operations, or safeguard your brand’s integrity, understanding an AI’s decision-making in a risk-free, transparent setting is invaluable. The live experiment at firmulate.com/live.html provides a rare window into this future.

Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html

Powered by Thorsten Meyer AI