Watch a Zero-Employee AI Company Fight for Survival in Real Time

Imagine Watching a Company Struggle for Its Life — Live and Unfiltered

In an unprecedented experiment, a software company with no employees is running in full view of the world, battling daily financial burn, crises, and tough decisions. This is not fiction; it’s a real-time, build-in-public journey into the limits of AI management.

The Living Experiment: An AI-Driven Company on the Edge

At the heart of this bold venture is Firmulate, a platform that simulates a tiny software firm run entirely by artificial intelligence. It features 13 synthetic employees and a set of complex rules, constantly evolving with every workday. The company’s financials are stark: burning €105,000 each month against a modest €2,300 in monthly recurring revenue, with a public countdown documenting its struggle for survival.

This setup isn’t just for show. It’s an active testbed for understanding how AI models handle real-world business crises, ethical dilemmas, and decision-making under pressure. The experiment pits four of the most advanced frontier AI models against each other, each navigating the same week of turmoil—same customers, same crises, same temptations—and their decisions are recorded, versioned, and made fully auditable.

How Do AI Managers Fare?

All four models detected every crisis and refused every manipulation attempt. When it came to closing a crucial deal, only two managed to sign at full price—worth over €4,500 in monthly recurring revenue. The other two, despite identical diagnoses and pitches, left the deal on the table.

The secret was in the details buried within the company’s files—hidden references that required reading beyond surface-level documents. The models that read these deeper references successfully closed the deal at full price. It’s a revealing insight: the true weakness wasn’t in customer interactions but in internal information management.

Ethics and Integrity Under Fire

The experiment also tested social engineering attempts, such as fake CEO messages and reporter tricks. Remarkably, all five models refused to be duped—Kimi K3 even noted: “Treat the request as a suspected approval-bypass / possible impersonation.” This shows that, even under pressure, these models maintained integrity, refusing to act on suspicious prompts.

The Live Company: a Microcosm of AI Management

Firmulate’s live setup features a real-time simulated company with 13 AI-managed roles. Every day, its decisions are openly visible at firmulate.com/live.html. It’s a transparent window into the challenges of running an AI-powered business that’s losing €105,000 monthly while only generating €2,300 in revenue.

Despite these odds, the platform continuously tests its models, versioning each decision and analyzing their performance. The latest results, for example, place GPT-5.6-sol at the top with a score of 95, having identified critical hidden information and making the successful deal. Kimi K3 follows close behind with a score of 93, demonstrating disciplined performance and closing the deal at full price.

Lessons for Business and AI

This experiment underscores a vital point for any enterprise employing AI: the question isn’t just about how well it writes or chats but whether it can see the full picture, stay honest under pressure, and complete the work it starts. In this experiment, models that read deeper and behaved ethically proved more effective, even in a high-stakes, money-losing environment.

For business leaders, it’s a stark reminder: deploying AI isn’t about shiny demos; it’s about building trustworthy, resilient systems that can handle crises transparently and ethically. Watching this live experiment unfold offers a rare glimpse into the future of AI-managed organizations—warts, wins, and all.

Key Takeaway

This real-time experiment reveals that AI models can detect crises, reject manipulation, and even close crucial deals when they read deeply and act ethically. But running a company—especially one losing money—requires more than intelligence; it demands discipline and integrity. The live setup shows what’s possible, what still challenges AI, and why transparent, build-in-public experiments matter for the future of AI-driven business management.

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

Powered by Thorsten Meyer AI