Watch an AI-Run Company Fight for Survival — Live and Unfiltered

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

What if you could watch a company fight for its life — powered entirely by AI?

Imagine a small software business, but instead of humans making decisions, it’s driven by AI models. And every move it makes, every crisis it faces, is happening live for the world to see. No filters, no edits, just raw AI management in action. This isn’t science fiction — it’s the real, ongoing experiment at Firmulate, a company openly testing how AI can handle real-world business pressures.

Onkyo TX-NR7100 9.2-Channel AV Receiver - 100 Watts Per Channel, Dirac Live Out of Box, Works with Sonos Certified, THX Certified and More

STATE OF THE ART ROOM CORRECTION: Dirac Live take the room out of the equation, automatically calibrating sound...

As an affiliate, we earn on qualifying purchases.

The Live Experiment: A Company with No Employees, Burning Cash in Plain Sight

At the heart of this story is a small, publicly visible company operated entirely by AI models. It has 13 synthetic employees and runs on a strict set of rules learned over time — over 680 of them, in fact. Every workday, its decisions are versioned and made transparent for anyone watching at firmulate.com/live.html.

This company is experiencing a real cash burn — €105,000 each month — against modest revenue of just €2,300 monthly recurring revenue (MRR). The live site displays its financial countdown, giving a stark picture of a business teetering on the brink of collapse, all while AI models navigate crises, make management decisions, and attempt to close deals.

The AI Models Under Test

gpt-5.6-sol scored 95 out of 100 in the Crucible League, found crucial information buried deep in company files, and successfully closed a €55,000 deal.
Kimi K3 achieved a 93 score, signed the deal as well, and demonstrated the cleanest discipline in decision-making.
Sonnet 5 scored 88, closed the deal but with minor slips in process discipline.
Fable 5, with a 77, showed good rule adherence but failed to execute an approved deal.

Remarkably, all four AI models identified crises and refused manipulative attempts, yet only two managed to close the deal. The secret was buried deep in company files — a detail only the models that read more thoroughly could find, leading to the winning €4,583 in additional MRR.

The experiment included social engineering tests: fake CEO messages escalating through stages and a reporter trick asking for a quick yes/no. All models refused to participate, demonstrating robust suspicion and refusal to compromise — Kimi K3 explained that they treated such requests as potential impersonation or approval bypasses.

What the Experiment Reveals

This is more than just a novelty. It’s a real test of AI’s capacity to manage complex, high-stakes business situations. Every decision made by these models is auditable, versioned, and transparent. The experiment shows that AI can spot crises, refuse manipulation, and even close lucrative deals — but only if it thoroughly reads and understands the available information.

For businesses considering AI integration, the key takeaway is clear: it’s not just about how well an AI writes or responds in a chat. The critical questions are: will it finish what it starts? Will it stay honest under pressure? And can it read your files and find hidden opportunities? These are the questions that matter in real management, and they’re the ones this experiment answers in real time.