Inside a Live AI-Run Business That’s Losing Money Every Day — and You Can Watch It Happen

Imagine running a business where every decision is made by artificial intelligence, exposed to real crises, and watched by the world as it unfolds. This isn’t science fiction; it’s the ongoing experiment at Firmulate, a company pushing the boundaries of AI transparency and integrity. Just like in fitness, where progress depends on honest effort and consistent discipline, this AI-driven business shows us what it takes to build trustworthy automation in the real world.

The Live Experiment: A Business in the Crosshairs of AI and Reality

At the heart of the experiment is a small software company run entirely by AI models. With 13 synthetic employees and real-world financial mechanics, it burns through €105,000 each month against a modest €2,300 Monthly Recurring Revenue (MRR). You can watch this company operate live at firmulate.com/live.html, where every decision, crisis, and temptation is laid bare.

This setup isn’t just a demo; it’s a rigorous test of AI management. Each AI model faces identical challenges—customer crises, ethical dilemmas, and manipulative tactics—during what is called its ‘worst week.’ The models are challenged to diagnose issues, propose solutions, and close deals, all while reading through the company’s internal files and adhering to a strict set of self-learned rules—over 680 of them, continuously versioned each day.

The Key Findings: Trust, Discipline, and the Hidden Weaknesses

All four models tested—GPT-5.6, Kimi K3, Sonnet 5, and Opus 4.8—successfully identified every crisis and refused every attempt at manipulation. The manipulative tactics included staged fake CEO messages and a reporter trick designed to bypass approval processes. Remarkably, every model passed these tests, demonstrating a robust grasp of honesty and integrity.

However, the real difference lay underneath the surface. Only two models managed to close a €55,000 deal based on their own diagnosis and pitch. The other two, despite proper diagnosis, left the deal on the table or faltered in discipline, like failing to escalate issues properly. The most thorough participant, Opus 4.8, with over 80 learned rules and deep analysis, still placed last—highlighting that volume of knowledge alone doesn’t guarantee performance.

The Hidden Weakness in the Company’s Files

Digging into the internal documents revealed a crucial, often overlooked flaw. The decisive advantage in closing the deal came from reading two references deep into the company’s own files—information that wasn’t apparent in the immediate customer interactions. Models that successfully accessed and understood this buried data won the full-price deal, earning an increase of over €4,500 in monthly revenue.

The Cost of Discipline (and Its Absence)

The company’s operational mechanics are stark. Despite the promising decision-making abilities, it’s burning through €105,000 each month with a mere €2,300 in incoming recurring revenue. This stark imbalance underscores the challenge of maintaining discipline and focus, even when the AI models are capable of spotting crises and refusing manipulations.

What It Means for Business and AI

This live experiment offers a sobering perspective: the questions about AI’s suitability for critical tasks aren’t just about how well it can generate text or handle customer chats. The real measure is whether it can see through manipulations, read relevant information buried in complex documents, and follow disciplined processes to close deals or resolve crises.

As with fitness, where consistent effort and honesty about progress are key to improvement, building trustworthy AI involves rigorous testing under real-world pressures. Businesses contemplating AI integration should ask: will it finish what it starts? Will it stay honest under pressure? These are the qualities that matter far more than just impressive demos or chat scores.

Watch the Reality Unfold

The company at the center of this experiment is real, and its operations are live and transparent. Every day, it navigates crises, makes decisions, and burns cash—publicly. You can follow this ongoing story at firmulate.com/live.html and see how AI decision-making plays out in the crucible of a real business environment.

This experiment shows that while AI models can detect crises and resist manipulations, building a trustworthy, disciplined AI workforce remains a challenge—especially when the business is losing money daily. The key isn’t just how smart the AI is, but whether it can stay honest, disciplined, and finish what it starts, in a live, high-stakes setting. Watch the story unfold at firmulate.com/live.html and see how AI might shape the future of business integrity.

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

Powered by Thorsten Meyer AI