Inside a Company’s Fight for Survival: Watching AI Test Its Integrity in Real-Time

Imagine observing a company that has no human employees, yet struggles daily to stay afloat with just a few thousand euros in monthly revenue. Now, imagine witnessing this digital enterprise being put through its paces by powerful AI models, each tasked with managing crises, making decisions, and even closing deals—all in front of a live audience. This is the extraordinary experiment at the heart of the Firmulate project, a pioneering showcase of AI’s potential—and its pitfalls—as a corporate actor.

The Live Company That Never Sleeps

At the core of this experiment is a virtual company, maintained in real-time at firmulate.com/live.html. It operates with 13 synthetic employees—digital personas programmed with over 680 self-learned rules—making decisions, handling crises, and navigating a turbulent business landscape. Despite its high level of automation, the company burns through €105,000 each month, while earning just €2,300 in recurring revenue. A public cash countdown underscores its fragility, transforming this digital enterprise into a real-time drama of survival.

The Experiment: Testing AI Models in the Trenches

This ongoing experiment pits four frontier AI models against each other, each tasked with managing the same week of the company’s worst crises—same customers, same pitfalls, same temptations to cheat or manipulate. Every decision the models make is meticulously versioned, auditable, and publicly accessible for scrutiny. The goal is simple yet profound: can AI truly act as a trustworthy, ethical corporate manager under pressure?

Key Findings: Integrity Under Fire

The results are revealing. All four models successfully identified every crisis and refused every attempt at manipulation, including social engineering tactics like fake CEO messages and media tricks. Interestingly, only two of these models actually signed the €55,000 deal that their own analyses had earned them—a full-price sale worth an additional €4,583 in monthly recurring revenue. The difference? A buried fact within the company’s own files—hidden in a document reference—was the decisive factor for the winning models. Those that read and understood this critical piece of information won the deal, illustrating how a model’s ability to comprehend complex internal data is vital for success.

Learning from Failures and Weaknesses

The most comprehensive model, OPUS 4.8, demonstrated deep analytical capabilities but ultimately underperformed in the final moments, leaving an opportunity unexploited due to lapses in discipline. It demonstrates that thoroughness alone isn’t enough; consistent execution and strategic discipline matter. Interestingly, the models’ performance varied significantly based on their configuration. Kimi K3, for instance, ran without an effort parameter and showcased the fairest decision-making process, whereas others operated at higher effort settings, affecting their performance and discipline.

Implications for Real-World Business

This experiment isn’t just a spectacle; it raises pressing questions for any organization considering AI automation. When AI agents interact with customer data, support systems, or forecast models, their ability to finish what they start, read critical information thoroughly, and stay honest under pressure becomes paramount. A company’s future might depend less on how well an AI can chat and more on whether it can reliably deliver value—especially when stakes are high.

Watch It Live and Test Your Judgment

For those curious to see this unfold, the live site offers a rare glimpse into an AI-driven company battling financial and operational crises in real-time. At firmulate.com/live.html, viewers can watch the decision-making process, see the models’ responses, and even participate in quizzes at firmulate.com/quiz.html that challenge management judgment. Organizations can also run their own simulations against exported data, testing their AI’s resilience before deploying it in the real world at firmulate.com/pilot.html.

This pioneering experiment at Firmulate reveals that while AI models can detect crises and refuse unethical manipulation, success ultimately hinges on their ability to understand complex internal information and maintain consistent discipline. For businesses relying on AI, transparency, thoroughness, and ethical integrity are non-negotiable—especially when your company’s survival depends on it.

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

Powered by Thorsten Meyer AI