Inside the World’s Most Transparent AI-Run Business Facing Daily Financial Collapse

What if you could watch a real company fight for survival, controlled not by humans but by AI models that are constantly tested and verified?

Imagine an ongoing experiment where a fully operational, yet entirely synthetic company navigates the worst week of its existence—every crisis, decision, and temptation recorded, auditable, and open for scrutiny. This is not a hypothetical or a simulation; it is the live demonstration of Firmulate, a pioneering platform that runs AI as if it were a real business, exposing both its strengths and weaknesses in stark detail.

The Unique Experiment: AI as Company CEO

Firmulate runs four different AI models—each with its own decision-making style—through the same relentless test: managing a small software company under duress. The models face genuine crises, customer pressures, and ethical dilemmas, all while their every move is versioned and stored for review. The goal? To assess whether these models can truly run a business and, more critically, whether they can do so honestly and reliably.

The Surprising Results

• All four models identified every crisis and refused to manipulate or bypass controls, demonstrating a high level of integrity.
• Only two of the four companies managed to determine and sign the €55,000 deal their own analysis justified, showing that decision quality can vary even among top-performing AI models.
• The critical weakness was buried within the company’s own files, not visible in customer interactions. When models read deeper into these internal documents, they secured a full-price deal worth over €4,583 in monthly recurring revenue (MRR).

Built-in Transparency and Trust

This experiment is designed to be entirely public, with every version of decision made available for review. It shows AI’s potential to uphold honesty and thoroughness in complex business scenarios, especially when the stakes are high. For instance, all models rejected attempts at social engineering, including staged CEO messages and reporter tricks. Kimi K3, one of the tested models, explained its reasoning clearly: “Treat the request as a suspected approval-bypass / possible impersonation.”

The Reality Behind the Curtain

The firmulate.com/live site offers a live window into this ongoing experiment. Today, a company with 13 synthetic employees is burning €105,000 each month against a meager €2,300 in monthly recurring revenue, with a public countdown to its cash exhaustion. Every weekday, new decision versions and findings are published, making this the most extreme example of building in public in the AI business world.

Lessons from the AI-Driven Business

While the models excelled at crisis detection and honesty, actual deal closure varied. The most thorough participant, OPUS 4.8, with over 80 rules learned and deep analysis, still left opportunities unexploited—highlighting that even the most disciplined AI can falter without careful management. Interestingly, the models’ performance differed slightly depending on the default effort parameters, with some running at default and others at high effort levels.

Why It Matters for Cloud and Infrastructure

This experiment is more than a curiosity; it’s a real-world benchmark for AI’s potential in managing critical business functions—whether in cloud infrastructure, hosting, or enterprise services. As AI models increasingly interact with your systems, the key questions shift from “Does it write well?” to “Will it finish what it starts?” and “Will it stay honest under pressure?”

The Future of AI in Business

Firmulate’s open, auditable approach gives cloud and infrastructure providers a glimpse of how AI can be tested and trusted before deployment. Running these models in a controlled, transparent environment can reveal vulnerabilities and ensure that AI-driven decisions meet the highest standards of integrity and effectiveness.

Next Steps and How to Watch

The live site at firmulate.com/live.html offers ongoing access to these experiments. For those interested in the decision dynamics, plain-language findings, or even testing their own management decisions, the platform provides a unique, real-time view of AI’s capabilities and limitations in a business setting.

Firmulate’s live experiment shows that AI can identify crises, refuse manipulation, and even close deals—if managed correctly. For cloud and infrastructure providers, it highlights the importance of transparency and rigorous testing before trusting AI with critical operations.

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

Powered by Thorsten Meyer AI