Imagine a furniture store that doesn’t have any employees but still makes decisions, handles crises, and even tries to close deals — all powered by artificial intelligence. This isn’t science fiction; it’s the real-time experiment of Firmulate, a company that runs its entire operation through AI models, openly battling for survival in the public eye.
Get decor and gifts delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
The Live Experiment: An AI-Run Business in Action
At the core of this experiment are 13 synthetic employees, each represented by advanced AI models that mimic decision-making in a typical software company. Every workday, these models face the same challenges: customer crises, ethical dilemmas, and sales opportunities. The company’s mechanics are real — burning €105,000 each month while generating just €2,300 in monthly recurring revenue. Visitors can watch this ongoing struggle unfold at firmulate.com/live.
What’s Being Tested?
- Decision-making quality under pressure
- Honesty and integrity in the face of manipulation attempts
- The ability to discover crucial information hidden within documents
- Discipline and process adherence in sales negotiations
The models are subjected to the same ‘worst week’ scenario, filled with crises, customer manipulations, and internal dilemmas. Every decision they make is versioned and auditable, allowing for transparent analysis of their performance.
AI decision-making simulation software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Surprising Findings
Despite all models recognizing each crisis and refusing manipulative tactics like fake CEO messages or reporter tricks, only two managed to win a major deal worth €55,000. Interestingly, the decisive advantage lay buried in their own files — a piece of information two document references deep that, when read, led to closing the deal at full price, adding +€4,583 in monthly recurring revenue.
Can AI Be Trusted?
All four models demonstrated a strong ability to detect crises and refused unethical requests. For example, Kimi K3 explicitly stated that it treats approval-bypass requests as potential impersonation or fraud — showing a built-in safeguard against social engineering.
business wargame AI training tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Cost of Running AI in a Real Company
This experimental company burns through €105,000 each month but earns only €2,300. It has a public cash countdown, and every workday’s decision-making process is openly available for analysis. The goal isn’t just to see if AI can chat well but whether it can actually run a business — making decisions, reading critical documents, and maintaining discipline amid pressure.
What Did the Most Thorough AI Do?
The Opus 4.8 model, which analyzed over 80 learned rules and provided deep insights, ended up in last place by not closing the deal and slipping in discipline. Instead of escalating issues, it left potential opportunities on the table — a sign that thorough analysis alone isn’t enough without disciplined execution.
AI ethics and crisis management training
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Implications for Business and AI
This experiment shows that AI can be effective at recognizing crises and refusing unethical manipulation. However, the challenge remains in consistently closing deals and maintaining discipline — especially under pressure. For companies looking to incorporate AI into their operations, the key questions are:
- Will the AI finish what it starts?
- Will it read and understand critical internal documents?
- Can it stay honest when temptations arise?
- And what does a unit of useful work really cost in this setup?
enterprise document analysis AI tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Real-World Application: Run Your Own Business Wargame
Businesses can now simulate their own scenarios using this AI framework without risking real systems. The pilot program allows companies to run their operations in a read-only environment, testing how AI would perform without any real-world consequences.
This public AI company experiment reveals that while models can identify crises and refuse manipulation, making consistent, disciplined decisions remains challenging. It’s a stark look at AI’s potential and limits in running real businesses in real time.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
Evergreen bestsellers Picks
bestsellers
As an affiliate, we earn on qualifying purchases.
