Imagine a fashion brand that operates entirely without employees, making decisions in real time, and showing every mistake and victory publicly. It sounds like a sci-fi fantasy, but it’s happening now. A pioneering experiment is putting AI models through their paces by running a virtual company in front of the world, revealing how machines handle crises, trust, and the messy reality of business — all live.
The Live Experiment: An AI Company Under Pressure
At the heart of this experiment is a small, simulated software company operated entirely by AI models. This company faces the same crises, customer issues, and temptations that real businesses encounter — but it’s all happening in a public, transparent environment at firmulate.com/live. The company has 13 synthetic employees, and every decision they make is versioned and auditable, providing a clear window into how each AI model behaves under pressure.
As an affiliate, we earn on qualifying purchases.
The Models in Competition
Four advanced AI models were tested — including GPT-5.6-sol, Kimi K3, Sonnet 5, and Fable 5. Each was tasked with guiding the company through its worst week, with identical scenarios, crises, and temptations. The results are revealing: all four models identified every crisis and refused to be manipulated. Yet, only two managed to close a critical €55,000 deal, earning full marks for their analysis and decision-making.
Interestingly, the decisive advantage was hidden in the company’s own files. The models that read and understood these internal documents secured the deal at full price, worth +€4,583 in monthly recurring revenue (MRR). Meanwhile, models that missed this buried fact failed to close the sale — a stark reminder that thorough reading and understanding are vital for success.
Trust and Ethical Boundaries
The experiment also tested how AI handles social engineering. Fake CEO messages and reporter tricks were used to try to manipulate the models. All five models refused to be duped, with Kimi K3 explicitly reasoning that such requests could be impersonation or approval-bypass attempts. This demonstrates that even in the face of pressure, these virtual employees maintain integrity — a crucial trait for real-world applications.
The Reality of a Digital Workforce
The live company isn’t just a demo; it’s a small operation losing €105,000 each month against just €2,300 in monthly revenue. It’s built on a set of over 680 self-learned rules, with every workday version being publicly updated. The goal? To understand whether AI can actually run a business — not just chat, but deliver measurable work.
In this ongoing experiment, the most thorough participant, Opus 4.8, scored the lowest — leaving opportunities unexploited and failing to escalate critical decisions. Meanwhile, Kimi K3, which ran without effort parameters, scored highly for fairness and discipline, showing that strict adherence to protocols is key. The results underscore a vital point for real companies: AI must be disciplined, transparent, and thorough to succeed in complex tasks.
Why This Matters for Business and Beyond
Today, AI is often showcased through impressive demos and chatbots. But the real measure of its value is whether it can complete complex, high-stakes work reliably and ethically. This experiment exposes the hidden weaknesses and strengths of AI decision-making in a live environment — revealing that trustworthiness, diligent reading, and discipline are what separate successful AI from mere gimmicks.
For industries like fashion and retail, where trust, authenticity, and decision accuracy matter deeply, understanding how AI handles crises and ethical dilemmas could reshape how brands operate in the future. Would your brand trust an AI to handle sensitive customer complaints or negotiate deals? The answer depends on whether it can read your files, stay honest under pressure, and deliver real results.
Watch, Learn, and Prepare
As this open experiment continues, viewers can see every decision, every crisis, and every outcome unfold. It’s a rare behind-the-scenes look at what AI-driven management could look like — and what it takes to build trustworthy, effective virtual teams. The implications stretch far beyond software companies, hinting at how AI might soon become a standard part of your business toolkit.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
