Imagine a Company That Runs Itself — and Is Barely Holding On
What if you could see an entire business operate daily, with no human employees, yet still face real crises, tough decisions, and the constant threat of failure? Welcome to the world of Firmulate, a live experiment that reveals what AI-managed companies look like in their most raw, unfiltered form.
Business Intelligence in the Age of AI: Modern Data Warehousing, Analytics and AI-driven Decision-Making
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Live Company: An Unconventional Business in Action
Firmulate runs a small software company with a unique twist: it employs 13 synthetic ’employees’ made up of AI models. These models are tasked with managing a real company that burns through €105,000 every month but makes only €2,300 in recurring revenue. This stark economic reality sets the stage for an ongoing test of AI decision-making under pressure.
Every weekday, the company’s decisions are recorded and versioned, creating a transparent, auditable trail of choices. The entire setup is live, accessible at firmulate.com/live, allowing anyone to observe how AI models grapple with crises, customer interactions, and ethical dilemmas in real time.
The Experimental Setup and Key Findings
- The experiment pits four leading AI models—gpt-5.6-sol, Kimi K3, Sonnet 5, and Opus 4.8—against the same set of business crises and challenges during the company’s worst week.
- Each model’s decisions are locked, versioned, and publicly accessible, making their reasoning transparent and auditable.
- All models successfully identified every crisis and refused manipulative tactics such as social engineering—fake CEO messages and reporter tricks—demonstrating robust ethical boundaries.
- The critical weakness was hidden deep within the company’s own documentation. Models that uncovered and read this buried fact were able to close a €55,000 deal at full price—adding +€4,583 in monthly recurring revenue (MRR).
- Despite their performance, only two models managed to finalize the deal; the others identified it but failed to execute the closing steps.
- Interestingly, the models that ran at the default API effort parameter (like Kimi K3) performed comparably to those configured for high effort, indicating that effort settings influence discipline but not necessarily accuracy.
Why This Matters for Business and Tech
This experiment isn’t about chatty AI or slick demos. It’s about real work—decision-making under duress, ethical resistance, and the ability to find critical information buried within company files. In a world where AI may soon interact directly with your CRM, support systems, or forecasting tools, the question isn’t just how well it writes but whether it can finish what it starts, stay honest, and deliver tangible results.
For the creators and consumers of music, audio, and creative tech, this story offers a crucial insight: AI models’ true test isn’t in entertainment or conversation, but in making disciplined, honest decisions that protect value—especially when stakes are high and resources limited.
Interview with the MONSTER AI: A Conversation about Power, Truth, and the Future of Intelligence
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Human and Machine Dance: What’s Next?
The live experiment is ongoing, and the results challenge assumptions about AI’s potential in business. With a real company losing money every day but still functioning publicly, observers can witness firsthand how AI models handle complex decision environments.
As the leaderboard shows, the highest-scoring model (gpt-5.6-sol with 95 points) was able to uncover the buried fact and close the deal, demonstrating the potential for AI to be both insightful and effective—if guided correctly. Meanwhile, the lower-performing models highlight the importance of discipline, thoroughness, and understanding hidden information.
All decisions are openly available for review, allowing enterprises to run their own simulations at firmulate.com/pilot.html and see how their AI agents might fare in similar crises, without risking actual business operations.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
The Project Management AI Handbook: Leveraging Generative Tools in Waterfall and Agile Environments
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The AI Marketing Canvas, Second Edition: A Five-Step AI Plan for Marketers
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
