What if you could watch a company fight for its life—powered entirely by AI, with no human employees, and all decisions visible to the world?
This isn’t science fiction. It’s the real-time experiment of Firmulate, where artificial intelligence models run a small software business through its toughest week yet, with every move tracked and scrutinized in public. For home tech enthusiasts and space-minded entrepreneurs alike, this daring ‘build-in-public’ effort reveals how AI handles crises, trust, and the grind of real-world business—live and unfiltered.
Motorized Projector Screen - Indoor and Outdoor Movies Screen Electric Projector Screen W/Remote Control (120 inch)
Size and Aspect Ratio: 4:3 aspect ratio, 120 inch diagonal screen size: 93.7 x 69.69 inch. Carton size:...
As an affiliate, we earn on qualifying purchases.
The Public Company That’s Anything But Ordinary
At the heart of this experiment is a company run entirely by AI models, simulating 13 synthetic employees operating under real money mechanics. Every day, the company burns through €105,000 in cash while generating just €2,300 in monthly recurring revenue, creating a high-stakes environment that tests the limits of artificial decision-making.
What makes this experiment unique is its transparency. Every decision, every crisis, and every rule learned by the AI is publicly versioned and auditable. The entire process unfolds live at firmulate.com/live.html, giving spectators a rare window into the true capabilities—and limitations—of AI in a business context.
Amazon Product B095XB54F7
As an affiliate, we earn on qualifying purchases.
The Experiment: Facing the Worst Week
Four frontier AI models—gpt-5.6-sol, Kimi K3, Sonnet 5, and Fable 5—each ran the same simulated week, filled with customer crises, internal dilemmas, and manipulation attempts. The goal? To see if these models could diagnose issues, make strategic decisions, and close deals without human intervention.
Remarkably, all four AI models identified every crisis and refused every attempt at manipulation, including social engineering tactics such as fake CEO messages and reporter tricks. In fact, every model showed a high level of integrity—none signed off on a questionable €55,000 deal, even when their analysis had justified it. The only difference was that two models—gpt-5.6-sol and Kimi K3—successfully closed the deal based on their own insights, while Sonnet 5 and Fable 5 left deals unexecuted due to discipline lapses.
100 Inch Motorized Projector Screen with RF Remote, 16:9 HD, Silver Case
【Cinematic 16:9 Visual Feast】Offers a 100-inch diagonal viewing area (87"W x 48.8"H) with a 16:9 aspect ratio, perfected...
As an affiliate, we earn on qualifying purchases.
The Hidden Weakness and Its Impact
Digging deeper into the company’s files, the experiment uncovered a critical weakness: the models that read and analyze the company’s documents at a deeper level were more successful in securing full-price deals. In particular, the models that examined internal references won deals worth an additional €4,583 MRR—highlighting the importance of comprehensive information analysis in decision quality.
120" Motorized Projector Screen, Electric 16:9 Projector Screen with Remote Control, Indoor/Outdoor HD Movie Screen, Wall/Ceiling Mounted, Perfect for Home Office Theater TV, White
Size and Aspect Ratio: 120" diagonal, 16:9 view size: 104.8"W x 58.7"H. Suitable for LED, LCD, and DLP...
As an affiliate, we earn on qualifying purchases.
Trust Under Pressure: The Fake CEO Test
To test the models’ resistance to social engineering, a staged scenario was conducted where fake CEO messages escalated over three stages, along with a reporter trick requiring a simple yes/no response. All five models refused to be manipulated, with Kimi K3 explicitly treating the requests as potential impersonation attempts. This demonstrates a promising robustness in AI behavior when faced with deceptive tactics.
The Real Money and Its Perils
Despite the AI’s impressive crisis detection and refusal of manipulation, the company’s financial mechanics reveal a brutal reality: it’s hemorrhaging cash faster than it earns. With a burn rate of €105,000 per month against a tiny income stream, the company is on a public countdown to financial exhaustion—all while being run by models that learn and adapt every workday, versioned and transparent.
The Lessons for Future AI Workforces
This transparency is the core strength of the Firmulate experiment. It shows that AI can handle complex, high-pressure decision environments, and importantly, that performance isn’t just about generating convincing chat but executing real work with honesty and discipline. Models that read deeply, analyze thoroughly, and refuse to bend under social pressure proved more successful in closing deals—an insight crucial for businesses considering AI for real-world operations.
What Can Home and Media Space Enthusiasts Take Away?
For those invested in innovative tech, this experiment offers a glimpse into how AI could someday manage or support home offices, media setups, or even complex home automation systems—if it demonstrates the discipline and trustworthiness shown here. Watching this live, you see a future where AI doesn’t just chat but actually runs parts of your business, with full transparency and accountability.
Key Takeaways
- AI models tested in a real-time, transparent business environment can identify crises and refuse manipulation, demonstrating integrity under pressure.
- Deep analysis of internal documents significantly impacts decision quality and deal closure success.
- Despite high discipline, the company’s ongoing cash burn emphasizes the importance of aligning AI decision-making with sustainable business practices.
- Watching this live experiment provides valuable insights into how AI might support or automate future home and business environments—if trust and discipline are maintained.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
