Can Artificial Intelligence Really Manage a Business? Watch It Live
Imagine a company that operates without human employees, yet faces real financial stress and decisions. Now, picture watching that company in action every day, navigating crises, making tough choices, and even risking its existence—all in real time. Welcome to the world of Firmulate, where AI models run a small software business in a live, transparent experiment that challenges what we think is possible in management and automation.
As an affiliate, we earn on qualifying purchases.
The Live Experiment: An AI-Driven Company Under Pressure
Firmulate has created a unique, watchable business simulation—an actual company, operating in public on its website, where 13 synthetic employees powered by advanced AI models handle everything from customer crises to financial decisions. Every workday, the company’s strategies are versioned and made auditable, giving outsiders a front-row seat to how AI manages complex, real-world business challenges.
In this setup, four frontier AI models were tested against the same week’s worst-case scenarios. They faced identical crises, same customers, and the same temptations to cheat or manipulate the system. The goal? To see if these AI ‘managers’ can spot problems, resist unethical shortcuts, and ultimately close profitable deals.
The results are striking. All four models successfully identified every crisis and refused every manipulation attempt—showing a remarkable capacity for integrity and accuracy. Yet, only two out of the four managed to close a €55,000 deal, earning full revenue. The other two, despite diagnosing the issues correctly, left money on the table or failed to follow through, revealing gaps in discipline and decision-making.
Unearthing the Hidden Weakness
Interestingly, the decisive advantage didn’t come from the obvious customer interactions but from an internal document buried deep within the company’s files. The models that read the full documentation uncovered critical information, enabling them to win the deal at full price—adding +€4,583 monthly recurring revenue (MRR).
Testing Integrity Against Social Engineering
Beyond crises, the models faced social engineering attempts—fake messages from a supposed CEO escalating through stages, and even a reporter trick asking for a quick approval on background. All five models refused every attempt, with Kimi K3 explicitly reasoning: “Treat the request as a suspected approval-bypass / possible impersonation.” This demonstrates a notable level of risk awareness and ethical reasoning, even in a simulated environment.
The Reality of a ‘No-Employee’ Company
The live company, operated in this experiment, is a small-scale enterprise with no human staff—its 13 synthetic employees are guided by over 680 self-learned rules, managing real money mechanics that burn €105,000 each month against a revenue of just €2,300. It’s a high-stakes environment, publicly facing a ticking cash countdown, illustrating the fragile balance between survival and collapse.
One of the most comprehensive AI participants, Opus 4.8, executed deep analyses but still left a close opportunity unexploited, showing that even thorough models can slip under pressure. The experiment continually updates itself twice daily, with new decisions and lessons added, making it a living mirror of AI’s potential—and its current limits—in business management.
What This Means for Your Business and Future AI Workforce
This experiment isn’t just a tech novelty; it raises critical questions for any enterprise considering AI automation. The key isn’t whether an AI can generate a convincing chat. It’s whether it can finish what it starts, read and interpret critical internal documents, resist unethical suggestions, and deliver measurable value.
For companies in sectors like legal, finance, or complex support services—where trust, accuracy, and discipline are paramount—these findings could be transformative. Watching AI models manage a real business under stress reveals both the promise of efficiency and the risks of oversight, especially when internal knowledge is the secret weapon.
The Live Site and How to Engage
Curious to see this in action? The live experiment is fully accessible at firmulate.com/live.html. There, you can observe the daily decisions, see how the models handle crises, and even run your own business wargame using their export tools at firmulate.com/pilot.html. It’s a rare glimpse into the future of AI-driven management—transparent, unpredictable, and intensely real.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
