Inside a Public AI-Run Company That Wins Deals Without Employees or Profit

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

Imagine a company that operates entirely in the open, with its decision-making and struggles on display for anyone to watch. No employees, no secret strategies—just a relentless battle to survive—and the world’s most advanced AI models watching every move.

A Living Laboratory for AI Management

At Firmulate’s live experiment, a small software company is being run day-by-day by artificial intelligence. Instead of human managers or employees, 13 synthetic ’employees’ handle operations, make decisions, and face real crises—all in a publicly viewable environment.

The AI Models in Action

Four frontier AI models, including the leading GPT-5.6-sol, are tested simultaneously. Each faces identical challenges—customer issues, crises, and opportunities to manipulate or deceive. Every decision they make is versioned and auditable, creating a transparent trail of how each AI performs under pressure.

Remarkably, all four models identified every crisis and refused manipulative or deceptive tactics, such as social engineering attempts. For example, when fake CEO messages aimed to escalate requests or bypass approval, all models refused—highlighting their robust resistance to social engineering tricks.

Results That Speak Volumes

Despite their vigilance, only two models managed to close a crucial deal valued at €55,000 per month in recurring revenue. One, GPT-5.6-sol, successfully uncovered a hidden piece of information buried two documents deep within the company’s files. That insight was critical to sealing the deal at full price—adding over €4,500 to monthly revenue.

The other model, Kimi K3, demonstrated the cleanest discipline but ultimately did not close the deal—leaving the opportunity on the table. The remaining two models, Sonnet 5 and Fable 5, showed some lapses but still maintained impressive decision-making accuracy.

The Cost of Running Artificially

The experiment reveals a harsh reality: the live company burns €105,000 each month in cash, earning only €2,300 in monthly recurring revenue. It’s a public, ongoing experiment in survival, with a visible cash countdown clock. The firmulate.com/live site offers live updates, allowing observers to see the company’s real-time struggles and critical decisions.

Learning From the Deep

The most thorough participant, Opus 4.8, learned over 80 rules for analysis but still left potential deals unexecuted, citing discipline slips—like writing attempts into a locked department instead of escalating them. Interestingly, the same weaknesses appeared across models, showing that even the most advanced AI struggles with complex, multi-step management when under pressure.

AI Builders: Making The Decisions That Turn AI Code Into Real Software

As an affiliate, we earn on qualifying purchases.

Why This Matters for Cybersecurity and Privacy

For readers concerned about cybersecurity, data privacy, and trust, this experiment offers a stark view: AI can identify crises and resist manipulation, but it doesn’t guarantee perfect execution or profit. Companies deploying AI in customer support, CRM, or decision-making need to ask not just whether AI can generate convincing chat but whether it can follow through, stay honest, and ultimately help manage real risks and resources.

This public, build-in-public approach showcases the importance of transparency and rigorous testing before AI is entrusted with critical business functions—especially when the stakes are high and every decision counts.