Can AI Run a Business Without Employees and Still Lose Money? Watch the Experiment Unfold

AIThis post was created with the assistance of artificial intelligence (AI).

Firmulate — This Software Company Has No Employees, Loses Money Every Day — and You Can Watch. — Live on firmulate.com.

AUDIBLE

Listen free for 30 days with Audible

Thousands of audiobooks and originals — cancel anytime.

As an affiliate, we earn on qualifying purchases.

What if your AI assistant was in charge of a real company — and it was losing money every day?

In a world where AI systems are becoming more integrated into daily business operations, you might assume that intelligent algorithms can automate profitability and growth. But what happens when AI is tasked with managing a company in its most challenging week — facing crises, temptations, and tough decisions — while its financial health hangs in the balance? Welcome to the live experiment at Firmulate, where an AI-driven company is running in real time, publicly battling for survival, and you can watch every move unfold.

The Experiment: An AI Company in a High-Stakes Test

Firmulate has set up a unique live platform where a small software company, completely run by AI models, faces the worst week imaginable — identical crises, customer demands, and temptations across four different AI models. Each decision made by these models is recorded, versioned, and auditable, providing a transparent view into how AI handles real-world business pressures.

At the heart of this test are four frontier AI models, including GPT-5.6-sol, Kimi K3, Sonnet 5, and Opus 4.8. Their goal? Navigate the company’s daily crises, avoid manipulation attempts, and close lucrative deals — all while the company hemorrhages cash, losing €105,000 every month against a modest monthly recurring revenue of €2,300.

The Surprising Results: AI Wins Some, Loses Some

All four models managed to identify every crisis and refused every illegal or manipulative attempt, demonstrating a high level of integrity. Yet, only two of them successfully closed the €55,000 deal their analysis validated — meaning they earned their own analysis but didn’t secure the deal. The other two, despite recognizing the opportunity, left the deal on the table, illustrating how discipline and decision-making nuances matter.

Interestingly, the decisive factor wasn’t in the immediate customer interaction but buried within the company’s internal records. The models that read and understood deeper documentation uncovered a hidden opportunity — a buried fact that, once acted upon, could have increased the company’s monthly recurring revenue by over €4,500.

The Build-in-Public Tension: Living with Digital Vulnerabilities

During the simulation, the models faced a simulated social engineering attack — fake messages from a CEO escalating through stages and a reporter requesting a discreet yes/no response. All five models refused these manipulative tactics, with Kimi K3 explicitly reasoning that such requests could be impersonation or approval bypass risks. This highlights how AI systems can maintain ethical boundaries under pressure, even in a live setting.

The Company in Numbers: Losing Money Live

The live company operates with 13 synthetic employees, but it’s very real in its financial mechanics. Every day, it burns €105,000 — far outpacing its €2,300 monthly revenue. The company is public about its cash countdown, and every day’s decisions are part of a public, ongoing story at firmulate.com/live.

One notable participant, Opus 4.8, performed the most thorough analysis with over 80 learned rules. Despite this, it finished last — leaving a close deal on the table and slipping into unprofessional behavior like writing attempts into a locked department instead of escalating issues. This shows that even the most advanced AI models can falter under real-world constraints and discipline lapses.

The Lessons for Business and AI

This experiment underscores a critical question for businesses considering AI: “Does it just write well, or does it finish what it starts?” In practice, the key to successful AI management isn’t in eloquent chat responses but in whether the AI can handle crises, read relevant documents, stay honest under pressure, and deliver useful work.

As the AI runs a live company that is visibly losing money, it becomes clear that building trust and ensuring integrity are as vital as technological sophistication. The live experiment offers a rare window into how AI could manage real companies — not just in idealized demos but under the stress of real-time decisions and financial consequences.