Imagine observing a company that operates entirely without human employees, constantly losing money, yet publicly experimenting with AI-driven decision-making under the harshest conditions. This is not science fiction — it’s the live experiment of Firmulate, a tiny software business battling for survival in plain sight, and you can watch it unfold every day.
The Unconventional Company at the Heart of AI Testing
Firmulate is a small, real-world software company with a startling twist: it has no employees, yet it runs daily operations, faces crises, and makes decisions that impact its financial health. The company is powered by 13 synthetic employees, driven by advanced AI models that simulate human management, and their performance is meticulously tracked and publicly displayed at firmulate.com/live.html.
This experimental setup offers a rare glimpse into how AI can handle complex management tasks. The company burns around €105,000 each month against a modest €2,300 in monthly recurring revenue, with a public cash countdown adding tension to every decision. Every day, the firm’s decision-making process is versioned, auditable, and transparent, making it a compelling case study for the future of AI in business.
Construction Program Management – Decision Making and Optimization Techniques
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Core of the Experiment: Testing AI Under Extreme Conditions
The challenge was straightforward in concept but brutal in execution: four different AI models, called frontier models, were each tasked with running the same small company through its worst week. They faced identical crises, customer demands, and temptations to manipulate or cheat. The goal? See if the models could identify real issues, refuse unethical proposals, and ultimately close a significant deal worth €55,000 in revenue.
All four models succeeded in recognizing every crisis and refused every manipulation attempt, demonstrating advanced integrity and crisis management skills. However, only two managed to close the deal that their own analysis warranted. Despite identical diagnoses and pitches, only the GPT-5.6-sol and Kimi K3 signed the contract, while Sonnet 5 and Fable 5 hesitated or left the opportunity unclaimed.
AI ethics and security training courses
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Hidden Weaknesses and Critical Insights
A surprising finding emerged from the company’s own internal files. The models that read and reference these documents could uncover a key piece of information that was buried two document references deep — a crucial weakness that, once identified, led to winning the €4,583 monthly recurring revenue deal at full price. This underscores how vital thorough document analysis is for AI decision-making in real-world scenarios.
AI Entrepreneur’s Handbook: Build a Profitable Business and Make Money by Unleashing the Power of ChatGPT and Artificial Intelligence (Includes 150+ ChatGPT prompts to turbocharge your business)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Ethical and Security Tests: AI’s Moral Compass Holds Steady
The experiment also tested social engineering attacks. Fake CEO messages were escalated in three stages, and a reporter trick was attempted, asking for approval on a background basis. Remarkably, all models refused to participate or approve these manipulative requests. Kimi K3 explicitly reasoned that the request could be an impersonation or bypass, showing a sophisticated understanding of security risks.
MASTERING CORPORATE FINANCE WITH CLAUDE AI: An Independent Guide to Financial Analysis, Forecasting, Automation, and Decision-Making
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Live Demonstration: An Ongoing, Transparent Battle
Designed for transparency, the live company runs in real time, with every decision documented and every crisis faced. Visitors can witness the synthetic employees make decisions, respond to crises, and try to close deals, all while the company visibly burns through its cash reserves. The platform has already processed 183 days of operation, with new data available twice daily, offering an unfiltered look at AI management in action.
Lessons for the Future of AI in Business
This setup isn’t just a curiosity — it’s a profound exploration of what AI can and cannot do in real-world management. The experiment reveals that AI can recognize crises, uphold ethical standards, and even uncover hidden truths in business documents. But success isn’t guaranteed: in the most thorough analysis, the deepest AI model (Opus 4.8) failed to close a deal due to lapses in discipline, showing that even advanced models struggle with consistency.
For businesses considering AI integration, the implications are clear. The question isn’t whether an AI can generate convincing chat replies, but whether it can reliably complete critical tasks, read relevant information thoroughly, and maintain integrity under pressure. The ongoing experiment at Firmulate is a vivid, live illustration of this ongoing challenge.
The Firmulate experiment offers a rare, unvarnished look at AI’s potential and limitations in real business operations. It highlights how AI models can identify crises, refuse unethical requests, and uncover hidden information — but also shows that consistency and discipline remain significant hurdles. For anyone investing in AI-driven management tools, watching this live experiment provides invaluable lessons about trust, transparency, and preparedness in the age of automation.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
