Imagine a fashion boutique run entirely by AI—no human staff, just algorithms making decisions, managing crises, and closing deals. Now, imagine watching this company struggle, succeed, and sometimes falter in real time. It sounds like science fiction, but it’s exactly what a groundbreaking experiment by Firmulate offers: a transparent, live look at how AI manages a small company under pressure.
Watching AI in Action: The Live Business Experiment
At the heart of this story is a real, functioning software company, operated entirely by artificial intelligence models. Instead of human employees, the company features 13 synthetic ’employees’—AI models that handle everything from customer interactions to crisis management. Every decision, every move, every slip is documented and available for public view at firmulate.com/live.
This setup is part of a radical build-in-public experiment designed to test AI’s ability to run a business ethically, efficiently, and honestly. The models are not just chatting; they face real business crises, ethical dilemmas, and competitive pressures, all while being measured against each other in a public leaderboard.
The Challenge: The Worst Week in Business
Each AI model was challenged with the same difficult scenario: a simulated week filled with customer crises, tempting manipulations, and complex decision-making. The models had to diagnose issues, pitch solutions, and close deals—just like human managers—but under strict rules that kept every move auditable.
The goal was clear: see which AI could navigate the worst week possible and still deliver results—closing genuine deals at full price, resisting unethical shortcuts, and maintaining discipline.
Key Findings: Honesty, Diligence, and the Hidden Weakness
The experiment yielded revealing insights:
- All four models identified every crisis and refused every manipulation attempt, demonstrating a strong grasp of ethical boundaries.
- Only two models managed to sign the €55,000 deal that their own analysis had earned them, showing consistency in execution.
- The real differentiator was a hidden piece of information buried two documents deep within the company’s own files—something the models that read these files uncovered and acted upon, resulting in a full-price deal worth +€4,583 in monthly recurring revenue (MRR).
- The models refused social engineering tricks—fake CEO messages and reporter tricks—each time, reinforcing their resistance to manipulation.
The Reality of a Build-in-Public Business
The company itself is not just an AI simulation; it’s a real software operation with daily expenses and a public cash countdown. It burns €105,000 monthly against a modest €2,300 in MRR, illustrating the fragile economics of such an experiment.
Every day, the company’s decisions are versioned, and its rules are learned and refined—more than 680 in total. The process is transparent, with real-time updates and insights available to observers, including the decision-making patterns and weaknesses that emerge under pressure.
Lessons from the AI-Run Business
This experiment underscores several essential questions for the future of AI in business:
- Can AI sustain honest, disciplined decision-making when stakes are high?
- Will AI systems read and understand critical internal information that can make or break deals?
- Are AI models resilient against social engineering and manipulation tactics?
The answer, so far, is promising—except when discipline slips, as seen with Opus 4.8, the most thorough model that still left on the table a crucial deal due to a lapse in escalation discipline.
The Broader Implications: Trust and Transparency
What does this mean for your business or brand? The experiment shows that building AI systems capable of ethical, honest work is not just about chat quality or superficial performance. It’s about their ability to finish what they start, prioritize internal information, and resist manipulative tactics—core qualities for trustworthy AI.
As firms increasingly adopt AI for customer support, decision-making, or even creative processes, understanding how these models behave under stress becomes vital. Watching this experiment unfold offers a rare glimpse into AI’s potential and its current limitations.
Takeaway
In a world where AI increasingly touches every facet of business—from fashion to finance—seeing it run a real company live, fighting for survival, and sticking to its principles is both fascinating and instructive. The experiment at firmulate.com/live proves that building trustworthy AI requires transparency, discipline, and resilience—qualities that are still being tested in real time.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
