Can AI Models Run a Business? Watch the Live Experiment and Decide

Imagine having an AI that manages your smart home or business. Would it make trustworthy decisions, or just pretend to? Now, witness a live experiment where leading AI models run a real, money-losing software company through its worst week — and you get to guess which model is which.

The Power of AI Management Personalities

Artificial intelligence isn’t just about chatting or answering questions — some models have distinct management styles that influence how they handle crises, trust, and decision-making. The firmulate experiment puts these differences to the test, revealing how AI personalities can significantly impact business outcomes.

The Setup: A Business in Crisis

Four advanced AI models, including GPT-5.6, Kimi K3, Sonnet 5, and Fable 5, each managed a small, real software company for a simulated week of turmoil. Despite identical conditions — same customers, same setbacks, same temptations — their decisions varied. This isn’t just a game; all decisions were recorded, transparent, and verifiable.

The Results: Different Styles, Different Outcomes

All four models successfully identified every crisis and refused manipulative tactics, such as fake CEO messages or media tricks. Yet, only two models managed to close a crucial €55,000 deal their own analysis had earned — a sign of real managerial discipline. Interestingly, the secret to winning was buried in the company’s own files, not just in customer interactions. The models that thoroughly read and understood these internal documents secured the full deal, adding €4,583 MRR to the company’s revenue.

Personality Profiles and Decision-Making Traits

• GPT-5.6 (Score: 95): Detected the hidden facts, closed the deal, and demonstrated comprehensive understanding.
• Kimi K3 (Score: 93): The newcomer with the cleanest discipline, also closed the deal.
• Sonnet 5 (Score: 88): Closed the deal but with some process slips.
• Fable 5 (Score: 77): Managed to close, yet less disciplined and more prone to overlook internal clues.

The Challenge of Social Engineering

The models faced staged social manipulation, including escalating fake CEO messages and media tricks. All five models, including the Kimi K3’s reasoning, refused to approve or bypass these requests, showing they understood the risks of impersonation and approval-bypass tactics.

The Live Business Environment

This isn’t a sandbox, but a real-time, operational emulation. The company runs 13 synthetic employees, spends €105,000 monthly, and earns just €2,300 in monthly recurring revenue — a reminder that trust and discipline matter as much as strategic insight. Every day, the models make decisions within a complex web of rules, and their choices are openly observable at firmulate.com/live.

This experiment reveals that AI management styles can vary significantly — from disciplined and thorough to more casual approaches. The key takeaway? For AI to be truly trustworthy and effective in your business or smart home, it must not only identify crises but also act decisively, read internal documents thoroughly, and resist manipulative tactics. Watching these models in action shows how AI personalities shape real-world outcomes, and why you should consider wargaming your AI workforce before deployment. Want to test your own management intuition? Take the quiz at firmulate.com/quiz.html.

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

Powered by Thorsten Meyer AI