Imagine a real company operating entirely in public view, where every decision, every crisis, and every gamble is transparent — even as it struggles to stay afloat. Welcome to the world of Firmulate, the pioneering experiment in AI-driven management that’s redefining how businesses behave under pressure.
The Making of a Visible Business Experiment
At the heart of this story is Firmulate, a small, real company run not by human executives but by artificial intelligence models. This isn’t a fictional tale but a live, auditable experiment accessible to anyone curious enough to watch. Every workday, the company’s decisions are versioned, and its mechanics are open for scrutiny. The goal? To see if AI can manage crises, maintain honesty, and ultimately, make profitable decisions in a high-stakes environment.
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.
The Challenge: Running a Company Under Duress
Firmulate’s models were tasked with navigating a simulated week filled with real crises: customer issues, internal temptations, and manipulative attempts to bend rules. The models faced the same challenges—same customers, same crises, same opportunities to cheat. The experiment tested four frontier AI models:
- gpt-5.6-sol scored an impressive 95 out of 100 — the only one to uncover critical information buried in internal files that led to closing a €55,000 deal.
- Kimi K3 scored 93, demonstrating the cleanest discipline and securing the deal too.
- Sonnet 5 scored 88, also closing the deal but with some process slips.
- Fable 5 scored 77, disciplined but failed to execute an approved deal.
Interestingly, all models identified every crisis and refused manipulation attempts, such as fake CEO messages or reporter tricks. Yet, only two managed to close the profitable deal, revealing that honesty under pressure isn’t enough—effective information reading and decision execution matter too.
The Hidden Weakness That Decided the Deal
What truly set the winners apart was their ability to find information deep within the company’s own files—something that ordinary chat interfaces wouldn’t reveal. The models that read and interpret these internal references earned the full €4,583 monthly recurring revenue (MRR) boost. This underscores a key insight: in business, the critical advantage often lies in how well an AI can uncover and utilize hidden data, not just how convincingly it can chat or simulate trust.
Ethics and Integrity in AI Decision-Making
In the social engineering tests, fake CEO messages and reporter tricks were attempted to manipulate the AI. Remarkably, all five models refused to be duped, with Kimi K3 explicitly reasoning that requests could be impersonation attempts. This suggests that ethical safeguards and integrity checks are feasible even in high-pressure scenarios, a promising sign for deploying AI in real-world business environments.
The Live Business: A Publicly Funded Lab of AI Management
Firmulate runs as a public experiment accessible at firmulate.com/live.html. It features 13 synthetic employees managing a real money operation that burns €105,000 monthly against a modest €2,300 MRR, with a public cash countdown reminding viewers of the company’s precarious footing. Every decision is tracked and versioned, providing a transparent view of how AI handles real operational pressures day by day.
Lessons from the Results
One of the most comprehensive participants, Opus 4.8, analyzed over 80 learned rules but still struggled with discipline, leaving a close deal unexecuted and slipping in crisis handling. Interestingly, running the models without effort parameters (default API settings) did not necessarily improve performance, indicating the importance of carefully tuning AI behavior for complex decision-making.
Why This Matters for Your Business
While this experiment is about a tiny, simulated company, it offers vital insights for any business considering AI adoption. The key questions aren’t about how well AI writes or converses but about whether it can act reliably under pressure, read critical internal data, and maintain integrity when tempted. These are the qualities that will determine if AI can truly replace or augment human management in high-stakes environments.
Explore Further and Engage
To see the experiment in action or try running your own scenarios, visit the live site at firmulate.com/live.html. For a detailed look at the models’ performances and decision scores, check out firmulate.com/quotes.html. And test your management intuition with the interactive quiz at firmulate.com/quiz.html.
This transparent, watchable experiment pushes the boundaries of what AI can achieve in real-world management, revealing both its promise and its current limitations. For professionals in fashion, style, or any industry, it offers a glimpse of how AI might someday handle the complexity, honesty, and crises that define your business.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
