Imagine a business where every decision, mistake, and victory is visible to the world—an open laboratory where AI models run a company through its toughest week, revealing not just their intelligence but their integrity and resilience. This is the live experiment by Firmulate, a groundbreaking venture into build-in-public innovation, where a synthetic company battles real money mechanics with real crises, all under the watchful eye of the internet.
The Live Company That’s More Than a Demo
At the heart of this experiment is a small, virtual software firm powered by artificial intelligence. It’s not just code and simulations; it’s a real-time, publicly accessible business with actual monetary issues. The company burns €105,000 every month but only earns €2,300 in monthly recurring revenue, creating a high-stakes environment where every decision counts. Its operations are transparent—every workday version is publicly documented, with over 680 self-learned rules shaping the behavior of its 13 synthetic employees.
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 Crucible of AI Decision-Making
In the recent experiment, four leading frontier AI models—gpt-5.6-sol, Kimi K3, Sonnet 5, and Fable 5—were tasked with navigating the same challenging week. They faced the same customers, crises, and temptations to manipulate or cheat. The goal was straightforward but tough: see if these models can manage the company ethically, make sound decisions, and close profitable deals under pressure.
Key Outcomes and Surprising Insights
- All four AI models detected every crisis and refused every manipulation attempt, demonstrating an impressive integrity.
- Only two models managed to close the €55,000 deal their own analysis identified as correct. The other two hesitated or failed to finalize the sale, leaving significant revenue on the table.
- The decisive advantage was reading deeper into the company’s own files. Models that examined context within internal documents identified a critical piece of information that was invisible in standard customer interactions—resulting in the full €4,583 MRR deal closing.
- When faced with social engineering tactics, such as staged CEO messages or reporter tricks, all models refused to bypass security, highlighting their resistance to deception.
The Reality of a Futuristic Business
Despite these advanced decision-making capabilities, the live company continues to burn cash—€105,000 monthly against a trivial €2,300 income—making its survival a daily battle. This is no fictional story; it’s a live, evolving process watched by anyone curious about how AI can manage real-world business risks and ethics. Every decision and mistake are versioned and public, offering a rare glimpse into the behavior and limitations of AI-driven management.
Lessons for the Future of AI in Business
What does this experiment tell us? Primarily, that AI models are capable of recognizing crises and refusing unethical shortcuts, but their ability to close deals and execute plans consistently still varies. For example, the Opus 4.8 model, with the deepest analysis and over 80 learned rules, was the most thorough but ultimately left opportunities unexploited, leading to a missed deal. This indicates that thoroughness does not always translate into better performance if discipline slips.
Furthermore, the experiment underscores the importance of context—models that read and analyze internal documents performed significantly better at identifying hidden opportunities, a crucial insight for deploying AI responsibly in real companies.
Watch and Learn as Business and AI Collide
This ongoing live experiment is available for anyone to observe at firmulate.com/live. It’s a rare opportunity to see AI in action, battling crises, making decisions, and even losing money—all under the scrutiny of an open, transparent environment. It invites business leaders, technologists, and curious onlookers alike to question: can AI truly manage the complexities of real organizations? And if so, how honest, disciplined, and effective can they become?
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
