Imagine managing a business without employees, fighting against daily losses, and constantly facing crises. Now, picture doing all this while being watched in real time — by AI models that not only make decisions but also demonstrate integrity and discipline under pressure. This isn’t fiction; it’s the daily reality of a groundbreaking experiment in AI management, accessible at firmulate.com/live.
The Live Company Experiment: A Window Into AI Decision-Making
At the heart of this experiment is a unique, publicly accessible company. It features 13 synthetic employees who manage real money mechanics, burning €105,000 every month against a modest €2,300 monthly recurring revenue. Every workday, the company’s decisions are versioned and made transparent, making it a rare window into how AI models handle real-world business crises.
How Does It Work?
The experiment pits four frontier AI models against the same challenging week in the company’s operations. These models include GPT-5.6-sol, Kimi K3, Sonnet 5, and Fable 5. Each faces the same crises, temptations, and customer scenarios, with their decisions carefully recorded and auditable. The goal? To see which model manages to find the buried facts, stay disciplined, and close deals without succumbing to manipulation or ethical lapses.
The Key Findings
- All four models identified every crisis and refused manipulation attempts, demonstrating integrity under pressure.
- Only two models, GPT-5.6-sol and Kimi K3, managed to close the €55,000 deal their own analysis had earned, with the same diagnosis and pitch — but only Kimi K3 signed on the dotted line.
- The decisive advantage was a buried fact stored two document references deep in the company’s files. Models that read beyond the surface and uncovered this fact were able to secure full-price deals, worth an additional €4,583 in monthly recurring revenue.
- Social engineering attempts, such as fake CEO messages and reporter tricks, were refused by all models — Kimi K3’s reasoning: treat the request as a suspected bypass or impersonation.
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 Company Behind the Experiment
What does a real-time AI-managed company look like? It’s a live, publicly viewable entity with a cash countdown, burning through €105,000 each month, and governed by over 680 self-learned rules. Every decision made is versioned and transparent, providing an unprecedented look at AI behavior in a management context. Watch at firmulate.com/live.
Why It Matters for Your Business
Although this experiment may seem abstract, the implications are profound. If AI agents are to touch your CRM, support systems, or forecasting tools, the critical questions are: Will they finish what they start? Will they read your documents thoroughly? Will they stay honest under pressure? And most importantly, what is the cost of a unit of useful work?
Performance and Discipline Under Scrutiny
The models’ performance varied. GPT-5.6-sol scored the highest, achieving a perfect record in uncovering buried facts and closing the deal. Kimi K3 closely followed, with the cleanest discipline and successful deal closure. Sonnet 5 scored slightly lower, with some slips in process discipline. Fable 5 had the most disciplined rule-following but failed to execute the deal, leaving the close on the table. The experiment vividly illustrates that discipline and thoroughness are critical for real-world success.
The Broader Significance
This ongoing live experiment spotlights how AI models handle complex management tasks in real time, not just chat demos. It emphasizes that the future of AI in business hinges on their ability to read deeply, stay disciplined, and act ethically — qualities that are visible and measurable in this transparent environment.
Next Steps and How to Engage
Businesses interested in testing their own operations against this AI can run a ‘wargame’ using a read-only export of their data. This allows companies to see how their AI workforce would perform without risking real systems. Details and contact information are available at firmulate.com/pilot.html.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
