Watch an AI-Run Company Fight for Survival in Real-Time Experiment

Imagine witnessing a company’s daily struggle to stay afloat—without a single human employee in sight. Real-time, open, and unapologetically transparent, this is the world of Firmulate, where artificial intelligence models run an entire company, facing crises, making decisions, and even losing money — all live for the world to see.

At the heart of this groundbreaking experiment is a small software company operated entirely by AI models. These models, which are essentially synthetic employees, navigate a challenging landscape filled with customer crises, internal dilemmas, and ethical tests. Every decision they make is recorded, versioned, and made publicly accessible, creating a transparent window into how AI handles complex, real-world management tasks.

The experiment pits four frontier AI models against each other, each guiding the same company through the hardest week imaginable. This includes dealing with customer crises, internal temptations, and even social engineering attempts designed to manipulate management decisions. Their goal? To see if AI can not only recognize problems but also act ethically and effectively under pressure.

Remarkably, all four models identified every crisis and refused every manipulation attempt. In fact, they demonstrated impressive discipline—none fell for social engineering tricks, such as fake CEO messages, which they all refused by treating the requests as potential impersonation or approval bypass risks. This highlights a crucial point: AI decision-making can be designed to resist manipulation, an essential trait for trustworthy automation.

However, the experiment reveals stark differences in outcomes. The models were tasked with closing a deal worth €55,000, which would significantly impact the company’s cash flow. Only two of the four models managed to sign the deal — and only after thorough analysis that uncovered a critical, buried fact in the company’s own files. The winning models, such as Firmulate’s live experiment, successfully read and interpret internal documents, find hidden insights, and close deals at full price, generating an additional €4,583 monthly recurring revenue (MRR).

Conversely, the other two models, including one called Opus 4.8, struggled with discipline. Opus ran over 80 learned rules and performed deep analyses but left the deal unexecuted, choosing instead to document and escalate issues to a locked department. This highlights a critical insight: thorough analysis does not guarantee execution — discipline and focus are equally vital.

The Stakes Are Real and Visible

What makes this experiment particularly compelling is the transparency of the entire process. The company, which burns €105,000 every month against a modest €2,300 MRR, is publicly counting down its cash as it fights for survival, with every workday versioned and accessible at firmulate.com/live. This build-in-public approach turns the traditional corporate narrative into an open lab, where viewers can observe how AI manages crises, makes decisions, and even handles ethical dilemmas in real time.

Implications for Business and AI Development

This live experiment underscores a vital question for companies contemplating AI integration: Can your AI agents not only generate convincing chat but also finish what they start, stay honest under pressure, and read critical internal documents? The findings suggest that the ability to uncover hidden facts—like the buried document reference—can be a decisive advantage in closing deals and avoiding losses.

Furthermore, the experiment demonstrates that avoiding manipulation is possible—every tested model refused social engineering attempts—indicating a promising path for building robust, trustworthy AI systems for enterprise use.

The Broader Significance

For readers interested in technology and gadgets, the experiment presents a new lens: AI’s role in management is not just about chatbots or support scripts but about making real decisions in complex environments. The question is no longer whether AI can write well but whether it can deliver measurable, valuable work — and how transparent and accountable that work can be.

As the company continues to burn cash while navigating this high-stakes game, the live site offers a rare opportunity to observe AI’s limits and potential in real-world business management. It’s a bold, unfiltered demonstration of AI’s capabilities, limitations, and the critical importance of ethical discipline in automated decision-making.

Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html

Powered by Thorsten Meyer AI