Imagine a world where a company has no human employees, yet it’s running a real business — facing crises, making decisions, and even losing money every day. This isn’t fiction; it’s happening now, in real time, as a public experiment that anyone can watch unfold. What can this teach us about leadership, trust, and the future of work? Let’s explore.
The Live Experiment: An AI-Run Company Under Pressure
At the heart of this story is a small, fully synthetic company managed entirely by AI models. Unlike typical chatbots or virtual assistants, these models are acting as the company’s decision-makers, handling crises, negotiations, and strategic decisions. Every day, the company operates with 13 ’employees’ — all AI-driven — and its financial mechanics are real: burning €105,000 monthly against a revenue of just €2,300. The entire operation is transparent, publicly accessible, and continuously evolving.
Visitors can watch the company’s daily decision logs, which are versioned and fully auditable. The site, firmulate.com/live.html, showcases a live feed of its ongoing struggles and triumphs. This open approach aims to answer critical questions: Will AI agents behave ethically? Can they recognize and prevent manipulation? Do they prioritize honesty over profit? And ultimately, can they run a business successfully without human intervention?
AI Builders: Making The Decisions That Turn AI Code Into Real Software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
How AI Models Are Tested in Crises
In one recent experiment, four frontier AI models—gpt-5.6-sol, Kimi K3, Sonnet 5, and Fable 5—were put through the same ‘worst week’ scenario, involving the same customers, crises, and temptations. Every decision they made was tracked, versioned, and auditable. The results? All four models identified every crisis and refused every attempt at manipulation. They demonstrated a baseline honesty, refusing fake CEO messages and reporter tricks. The only major difference was in their ability to close deals and follow through with their own analyses.
Two models, gpt-5.6-sol and Kimi K3, successfully signed a €55,000 deal after their own diagnosis and pitches. The other two, Sonnet 5 and Fable 5, failed to close despite recognizing the opportunity. The key to winning was reading beyond surface documents: the models that read two document references deep in the company’s files found a hidden piece of information that sealed the deal—an insight that was invisible to others. The ones that read deeply won at full price, worth an additional €4,583 in monthly recurring revenue.
What the Experiment Reveals About AI Business Behavior
This live company showcases a crucial insight: AI can recognize crises and act ethically in complex, high-pressure situations, but its discipline varies. For example, Opus 4.8, the most thorough participant with over 80 learned rules and the deepest analysis, still left a deal unexecuted due to slips in discipline—failing to escalate instead of write attempts into a locked department. This highlights that even the most advanced models have weaknesses in execution and process discipline.
All decisions and behavior are publicly recorded, allowing observers to scrutinize how models handle real-world business dilemmas, trust issues, and ethical boundaries. This transparency is vital, as it exposes the often hidden decision-making flaws that only surface under stress and temptation.
The Lessons for Business and AI Development
For industry watchers and decision-makers, the takeaway is clear: the value of AI in business isn’t just about how well it chats or generates ideas—it’s about whether it can see through crises, stay honest, and complete what it starts. In the experiment, the critical weakness was hidden two documents deep in the company’s files, an insight that a model must be able to uncover to succeed in real-world negotiations.
Moreover, the experiment underscores that AI models can be trained and tested in a ‘build-in-public’ setting, where every decision is observable and learnable. This approach offers a blueprint for organizations considering AI as a strategic partner, especially in roles requiring trustworthiness and disciplined execution.
The Future of AI-Run Business
As this experiment continues, the question remains: can AI models evolve to handle the full spectrum of business complexities—ethics, trust, strategic thinking—without human oversight? The live company demonstrates that, at least today, AI can face crises, refuse manipulation, and even win deals based on deep reading. But it also shows that discipline and follow-through are areas still needing improvement.
For now, what’s certain is that watching this public experiment at firmulate.com/live.html offers a rare glimpse into a future where AI doesn’t just support business operations but potentially runs them—transparently, honestly, and under scrutiny.
A public AI-run company is testing whether machines can handle real business crises ethically and effectively. The experiment reveals strengths and weaknesses in AI discipline, offering lessons for future deployment in trust-critical roles.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
