Watch a Company That’s Going Broke Live—With AI as Its Workforce

What if your team was entirely made up of AI—and it was losing money every day?

Imagine watching a real company stumble through its worst week, with every decision, crisis, and temptation laid bare for the world to see. That’s exactly what the live experiment from Firmulate offers—a stark, unfiltered look at how artificial intelligence manages a business under pressure, and whether it can truly replace human judgment in the chaos of real-world decision-making.

The Reality of a Digital Company in Survival Mode

At the heart of this experiment is a small, real software company—its operations are entirely simulated, but its financial mechanics are real. It employs 13 synthetic employees, each driven by advanced AI models, and every day is a battle to stay afloat. With a monthly burn rate of €105,000 against a revenue of just €2,300, the company’s cash countdown is public and ongoing. Every decision made by these AI ’employees’ is versioned and auditable, providing a transparent window into their decision processes.

This setup isn’t just for show. It’s a serious test of AI’s ability to handle crises, navigate ethical dilemmas, and ultimately, close deals—like the €55,000 contract that only two of the AI models managed to sign out of four contenders.

The AI Models: Different Approaches, Same Challenges

The experiment pits four frontier AI models against each other in this high-stakes environment. These models, including the well-known GPT-5.6-sol and others, all faced the same intense week of operational crises, customer complaints, and ethical tests. Each decision was scrutinized, and every model was given the chance to find the buried facts in company documentation—an area where some models excelled, giving the company the edge to close a deal at full price.

Interestingly, the models were able to detect manipulative social engineering tactics—like staged CEO messages and reporter tricks—and refused to play along, demonstrating a grasp of trust and impersonation issues that are critical in real business scenarios.

The Live Experiment: A High-Stakes Game of Business

This isn’t just a theoretical exercise. The company runs every business day, and its progress is visible at firmulate.com/live.html. It’s a real-time, ongoing story of a business fighting for survival, with every decision, crisis, and ethical dilemma laid bare for observers. The goal isn’t just to prove AI can handle tasks but to measure whether it can consistently deliver useful, honest work under pressure.

In one notable case, the most thorough model—Opus 4.8—demonstrated exceptional analytical depth but failed to close the deal. It left opportunities on the table and slipped into internal conflicts, showing that even the best AI can struggle with discipline and strategic focus, especially under stress. Meanwhile, simpler models like Kimi K3 managed to close deals more cleanly, highlighting different strengths and weaknesses.

What This Means for the Future of AI in Business

This experiment underscores a crucial point: the ability of AI to not just generate convincing chat or support responses but to actually finish what they start, read crucial documents before acting, and remain honest under pressure. If AI tools are to touch your customer relationship management, support queues, or forecasting systems, these are the real questions to ask.

The performance league table—where GPT-5.6-sol leads, followed by Kimi K3, and then the Sonnet models—offers a tangible measure of progress. But the takeaway for business leaders is clear: AI’s value isn’t just about how well it talks; it’s about whether it can keep its promises and deliver real results in the face of adversity.

The Bottom Line: Building in Public, Building for Survival

This visible, ongoing experiment by Firmulate is a raw, unfiltered glimpse into a future where AI manages real companies, and those companies are fighting to even stay afloat. It’s a story of resilience, discipline, and the importance of uncovering hidden weaknesses before deploying AI at scale. Watching this live company struggle and adapt offers lessons as valuable as any financial report—and perhaps, a warning about overestimating what AI can achieve without careful oversight.

Key Takeaway

This live experiment shows that AI models can handle crises and ethical dilemmas, but their ability to close deals and stay disciplined under pressure varies. For businesses considering AI automation, it emphasizes the importance of transparency, trust, and thorough testing—before AI takes on mission-critical roles.

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

Powered by Thorsten Meyer AI