Watch a Live AI-Driven Business Struggle for Survival in Real Time

Imagine Watching a Business Fight for Its Life—Every Day

For travelers and outdoor enthusiasts, the thrill of facing the unknown is part of the adventure. Now, imagine witnessing a company navigate its own high-stakes survival story, live and unfiltered. That’s precisely what the experiment at Firmulate offers—a rare window into how artificial intelligence manages a business on the brink, day after day.

The Live Experiment: An AI Company Under Pressure

Far from being a typical tech demo, Firmulate’s venture is a real, functioning company that runs every business day with a twist: it’s entirely managed by AI models. These models act as 13 synthetic employees, tasked with running a small software firm facing the worst possible week. They handle real crises, real customer demands, and real financial mechanics, all while being monitored and analyzed in real time.

The company is burning through €105,000 each month but earns just €2,300 in monthly recurring revenue—an unsustainable situation. Yet, this isn’t a story of failure; it’s a meticulous experiment designed to test AI’s decision-making under stress, with every choice visible to the public at firmulate.com/live.

How Do AI Models Perform in Critical Moments?

Four state-of-the-art AI models, representing the frontier of language and decision-making, faced identical challenges. These included crises like customer issues and manipulative social engineering attempts. Remarkably, all four models identified every crisis and refused to be manipulated, showcasing a high level of integrity and awareness.

However, only two models managed to close a lucrative deal worth €55,000—full price—and this was purely based on their own analysis and decision. The other two, despite diagnosing the problem correctly and making the same pitch, failed to sign the deal, illustrating how subtle process slips can cost real money.

The Hidden Weakness: Deep Within the Files

One of the most revealing findings was that the decisive advantage in winning the deal lay buried two document references deep in the company’s internal files—not in the immediate customer interaction. Models that took the time to read and analyze these internal references secured the deal at full price, demonstrating how crucial deeper information processing is for success.

Defending Against Social Engineering

In a staged social engineering attack, a fake CEO message was escalated through multiple stages, and a reporter attempted to get a quick approval with a simple yes/no response. All five models tested refused to bypass security or impersonate authority, citing suspicion and the risk of impersonation. This shows that even sophisticated AI can be trained to resist manipulation attempts designed to exploit human-like trust.

The Reality of a Money-Losing Business

The live setup isn’t just an experiment—it’s a brutal portrait of a company in trouble. With over 680 self-learned rules and every decision versioned, it is a transparent glimpse into what building AI-driven management looks like in practice. Every workday’s actions are recorded, analyzed, and learnable, offering invaluable insights for those interested in deploying AI in real business environments.

The Lessons for the Future

This experiment underscores a critical point: the question isn’t whether AI can generate polished chat responses, but whether it can complete meaningful, honest work reliably under pressure. Can AI read relevant internal documents? Can it stay disciplined despite temptations? And can it produce measurable value for your enterprise?

While the models excelled at crisis detection and resisting manipulation, subtle process lapses—like not escalating issues properly—cost opportunities. The real takeaway is that AI management systems must be designed to handle complexity with discipline and honesty, especially when stakes are high.

Accessible, Transparent, and Real

Anyone interested in seeing this in action can visit the live experiment. Watch the business run daily, listen to the decisions, and even challenge the AI with a quiz at firmulate.com/quiz.html. For those contemplating AI adoption, it offers an unvarnished look at both its potential and its current limitations, in a setting that’s as transparent as it is compelling.

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

Powered by Thorsten Meyer AI