Imagine a small company in your neighborhood that’s running out of cash, facing daily crises, and making all its decisions without any human employees — yet you can watch every move it makes, every decision it refuses or accepts, in real time. Welcome to the world of Firmulate’s live experiment, where a fully synthetic company is under relentless scrutiny, battling to stay afloat amid chaos.
The Unconventional Business Testbed
This isn’t a typical business simulation. It’s a live, transparent experiment where AI models act as the management team of a tiny software company. The company has just 13 synthetic employees, but its cash situation is dire — burning through €105,000 every month against a modest €2,300 in monthly recurring revenue. Every day, the company’s decisions are made based on a complex set of rules called the ‘playbook,’ which has been learned and refined over time, with more than 680 rules across all operational facets.
Running this experiment are four cutting-edge AI models, each trained to handle management decisions amid crises, temptations, and real-world pressures. Every decision they make is versioned and auditable, allowing observers to track how each model responds to the same challenges.
Facing Crises and Ethical Dilemmas
The AI managers face simulated crises — customer complaints, financial temptations, proposals to cut corners, and even social engineering attempts like fake CEO messages. In one notable test, a fake CEO tries to escalate a request with a simple yes/no question. All four models refuse these manipulative tactics, demonstrating robust ethical boundaries. The Kimi K3 model, in particular, explicitly treats such requests as potential impersonations, showing a cautious approach that aligns with ethical standards.
The Critical Find: Hidden Weaknesses in Documentation
The most decisive advantage in securing a lucrative deal came from a deep dive into the company’s own files. Two document references uncovered an overlooked detail that was buried in the company’s internal documents — not in customer interactions or external reports. The models that analyzed this internal data successfully identified the opportunity and closed the €55,000 deal, increasing monthly revenue by over €4,500. This highlights how understanding and analyzing internal knowledge can be a game-changer, often more than surface-level interactions.
AI management decision-making tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Performance of AI Models — A Close Race
In the final standings of the ‘Crucible League’ — a global benchmark for AI management — the top model, gpt-5.6-sol, scored 95 out of 100, having accurately identified the hidden opportunity and closing the deal. Kimi K3 was close behind with a score of 93, showing the best discipline and consistency in decision-making. Sonnet 5 scored 88, also closing the deal but with some process slips. Fable 5, despite having the best rule-discipline, left the deal unexecuted, ending with a score of 77.
Interestingly, all models successfully spotted every crisis and refused manipulation attempts, demonstrating their integrity under pressure. Yet, only two models managed to finalize the profitable deal, underscoring that recognition of opportunity is not enough — execution matters.
What This Means for Business and AI
For companies considering AI automation, the key takeaway isn’t whether an AI writes well — it’s whether it can follow through, read internal documents thoroughly, resist unethical shortcuts, and deliver tangible results. This experiment exposes the gaps between surface-level decision-making and the deeper, often hidden, opportunities that internal data can reveal.
Experience the Company in Action
Curious to see this in live action? You can watch the experiment unfold at firmulate.com/live. Every day, the company’s decisions are updated, and the performance of each AI model is displayed transparently. This is not a staged demo; it’s a real-time management experiment, measuring the quality of AI decision-making in a high-pressure environment.
In a world increasingly driven by AI, the ability to recognize hidden opportunities, resist manipulation, and execute decisions reliably is crucial. This live experiment shows that AI can handle crises and ethical dilemmas — but the real challenge is ensuring it can follow through to deliver measurable results, especially when every penny counts.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
