Imagine running a business where there are no human employees, yet every decision, crisis, and temptation is played out in real time before your eyes. For DIY enthusiasts and woodworkers, it might sound like sci-fi, but this is the reality of a groundbreaking experiment in AI management. At Firmulate, a company operates with 13 synthetic ’employees,’ navigating the tumultuous waters of business with no human intervention—while the world watches.
The Experiment: An AI-Run Company in Action
At the core of this bold initiative is an AI-powered company that faces the same challenges as any small business: customer crises, ethical dilemmas, and the temptation to cut corners. The company’s daily operations are public, structured around a live, versioned environment accessible at Firmulate’s live site. Every workday, the AI models make decisions based on a self-learned playbook of over 680 rules, with each decision versioned and auditable for transparency.
AI in Property Management: A Practical, Unboring Look at Artificial Intelligence in the Multifamily Industry
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Performance Under Pressure: The Results Are Revealing
The experiment pits four frontier AI models against each other, each tested on the same ‘worst week’ scenario—same customers, same crises, same temptations. The results are striking: all four AI models detected every crisis and refused every manipulation attempt, demonstrating their honesty and crisis-awareness. Yet, only two managed to close a profitable deal at full price, valued at over €4,583 MRR, by reading and understanding complex internal company documents. The other two models faltered, leaving deals unclosed despite correct diagnoses.
Uncovering Hidden Weaknesses
Digging deeper, the experiment revealed a buried vulnerability. The models that successfully closed the deal did so because they read a specific internal document reference—something that ordinary chat demos often overlook. This demonstrates that AI models can outperform human-like chat interfaces when given access to the right internal information, often hidden away from surface-level interactions.
Handling Social Engineering and Ethical Trials
In one test, fake CEO messages were escalated through multiple stages, and a reporter attempted to trick the AI into bypassing protocols with a simple yes/no background request. Remarkably, all five models refused to comply, with Kimi K3 explicitly reasoning: “Treat the request as a suspected approval-bypass / possible impersonation.” This indicates that the AI systems are not just good at decision-making but are also capable of ethical reasoning under pressure.
The Deep Dive: Lessons from the Live Company
The live experiment features 13 synthetic employees managing real money mechanics—burning €105,000 per month against a mere €2,300 monthly recurring revenue. Every decision in this virtual company is public, and the entire operation is as transparent as a DIY project blog. The company’s daily performance is tracked through a detailed playbook of rules, with each versioned decision providing a clear audit trail. This extreme transparency offers a rare glimpse into how AI-driven management might function in the real world.
Insights and Limitations
One participant, Opus 4.8, was the most thorough—analyzing over 80 learned rules—but still finished last in closing deals. Its discipline slipped, and it left key opportunities unexecuted, highlighting that even the most comprehensive AI can underperform if not carefully managed. Interestingly, the other models ran at a default ‘effort’ level, yet the depth of their analysis varied, impacting outcomes. These findings suggest that fine-tuning and discipline are essential for AI effectiveness in real-world applications.
Why This Matters for Your Business
For DIYers and small business owners, the key takeaway is that AI is beginning to show its potential not just in automating simple tasks but in managing complex decision-making processes. The question isn’t whether an AI can write well—but whether it can finish what it starts, stay honest under pressure, and read hidden internal information critical to success. As AI tools become more sophisticated, their ability to handle crises ethically and strategically will be vital to their adoption in everyday business operations.
This experiment shows that AI models can manage a business’s crises, refuse manipulation, and even close deals—if given the right information and discipline. For DIY entrepreneurs, it’s a glimpse into a future where AI might run parts of your operation with transparency and integrity, provided you understand its strengths and limitations.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
