Inside a Money-Losing AI-Run Company That’s Publicly Testing Its Limits

Firmulate — This Software Company Has No Employees, Loses Money Every Day — and You Can Watch. — Live on firmulate.com.

Imagine a small business that operates in the open, with every decision, crisis, and failure played out before the world. A company so transparent that you can watch its daily struggles, see its cash burn, and witness its efforts to stay afloat—all without human employees. Welcome to the world of Firmulate, an experimental AI-powered company that’s redefining transparency in the business world.

The Live Experiment: An AI Company Under the Microscope

Firmulate runs its entire operation with 13 synthetic employees—AI models tasked with managing a small software firm facing daily crises. Every decision they make is recorded, versioned, and accessible for public scrutiny at firmulate.com/live. This setup isn’t just a demo; it’s a real-time battle to sustain a business that burns €105,000 monthly against a modest monthly revenue of €2,300. That’s a cash countdown that puts the company’s survival in stark perspective.

Testing AI Models in a High-Stakes Environment

Four advanced AI models, including GPT-5.6-sol and Kimi K3, were subjected to the same brutal week—a scenario filled with customer crises, internal dilemmas, and the temptations to manipulate the system. Each model’s decisions were not only made but also fully auditable, revealing their capacity to detect crises and resist unethical requests.

Remarkably, all four models identified every crisis and refused manipulative attempts, such as social engineering tricks like fake CEO messages. Yet, only two managed to close a critical deal worth €55,000—indicating that recognizing problems and resisting unethical pressure does not guarantee business success.

The Hidden Weakness in the Files

Deep within the company’s own documentation lay a decisive piece of information—two document references that, if read, could have secured a full-price deal, adding +€4,583 to monthly recurring revenue (MRR). The models that examined these files successfully closed the deal; those that missed them left the opportunity on the table.

Building in Public: The Extreme Transparency

This experiment is a build-in-public showcase of AI’s potential and limitations. Every decision, every slip, and every success is openly recorded, providing a rare view into how AI models perform under pressure and in complex, real-world scenarios. Watching this unfold at firmulate.com/live offers insights into whether AI can truly manage a business—honestly, effectively, and consistently.

Lessons from the Experiment

All tested models recognized crises and refused unethical suggestions, demonstrating strong ethical resistance.
Closing a deal required reading the right documents—a hidden weakness in the AI’s approach.
The most thorough model, Opus 4.8, showed the deepest analysis but still left opportunities unexploited, highlighting discipline as a key challenge.
Despite the AI’s capabilities, the company’s financials remain bleak, burning €105,000 per month with minimal income.

AI Builders: Making The Decisions That Turn AI Code Into Real Software

As an affiliate, we earn on qualifying purchases.

Implications for Business and AI

This experiment raises fundamental questions: Can AI models be trusted to run parts of a business? Will they stay honest under pressure? And crucially, will they be effective enough to justify their cost? For companies that consider deploying AI in customer support, sales, or management, these insights are vital. The question isn’t just whether the AI can generate good chat—it’s whether it can deliver consistent, honest, and comprehensive work that leads to real results.