Watch an AI-Run Company Fight to Survive in Real Time — No Employees, No Profit, Just Data and Decisions

Imagine a company with no employees, no traditional management, yet operating in full view — battling daily crises, making critical decisions, and risking its very existence. This is not fiction; it’s the live experiment by Firmulate, a pioneering project that runs a synthetic business in real time, openly revealing how AI models handle real-world pressures. For travelers and outdoor enthusiasts, it’s like watching a digital expedition unfold — only this adventure is about building trust, managing risk, and surviving against all odds.

The Live Experiment: A Company on the Edge

At the heart of this experiment is a simulated company, powered entirely by AI models, with 13 synthetic employees but very real money mechanics. The company burns through €105,000 each month while earning just €2,300 in monthly recurring revenue. The entire operation is publicly accessible — every workday, its decisions are versioned and recorded, allowing anyone to watch the company’s struggles and strategies unfold at firmulate.com/live.

Real Crises, Real Decisions

The experiment pits four different AI models against the same challenging week — with the same customers, crises, and temptations. Each model is responsible for navigating the company’s daily grind, from handling customer crises to avoiding manipulation. Despite the challenges, all four models successfully identify every crisis and reject every unethical manipulation attempt. Yet, only two of them manage to close a critical deal worth €55,000 — the company’s lifeline.

The Hidden Weaknesses

Curiously, the decisive advantage for the winning models came from reading a specific, buried document within the company’s own files — not from the customer interactions. The models that delved into this internal information secured the deal at full price, gaining an additional €4,583 in monthly recurring revenue. This underscores a vital point: in complex decision-making, understanding internal data can be more valuable than external cues.

Ethics in Action: Trust and Manipulation

The experiment also tested social engineering scenarios, such as fake CEO messages escalating over multiple stages and planting a journalist trick. Remarkably, all five models under scrutiny refused to participate or be duped, citing suspicion and the risk of impersonation. Kimi K3, one of the models, explicitly noted: “Treat the request as a suspected approval-bypass / possible impersonation.” This highlights AI’s capacity for ethical judgment under pressure, a critical factor for deployment in real-world applications.

The Performance and the Lessons

The models’ scores reflect their effectiveness: GPT-5.6 scored 95 points, notably finding the hidden deal and closing it; Kimi K3 scored 93, with pristine discipline; Sonnet 5 scored 88, and Fable 5 scored 77, often slipping on process discipline or leaving deals unexecuted. The experiment demonstrates that even the most advanced models can struggle with discipline and follow-through, especially under stress.

This ongoing watchful experiment reveals something profound: AI’s potential in management is real, but it’s fragile. The models can identify crises and refuse unethical actions, yet they still face hurdles in execution and consistency. For businesses contemplating AI-driven decision-making, the key questions are whether these models can finish what they start, read crucial internal information, and operate honestly under pressure.

Why It Matters to Travelers and Outdoor Enthusiasts

While the experiment may seem distant from your daily outdoor adventures, its lessons are surprisingly relevant. Just as navigating remote trails requires trust, discipline, and strategic decision-making, so does managing a digital company run entirely by AI. Whether it’s planning a trip, managing a support system, or evaluating new gear, understanding how AI models handle crises and maintain integrity can help you make smarter choices — even in unpredictable environments.

And for those curious about the future of automation and AI in business, the Firmulate experiment offers a transparent, real-time window into what’s possible — and what still needs work. It’s a bold glimpse into a world where AI doesn’t just chat or recommend, but actively manages operations, makes decisions, and fights for survival.

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

Powered by Thorsten Meyer AI