Imagine coaching your child through a tough week: Would you give them the same advice as your best friend, or tailor your guidance based on their personality? Now, what if the ‘child’ was an AI running a real company? This is exactly what a groundbreaking experiment by Firmulate is exploring—measuring how different AI models handle leadership, honesty, and crisis management in a simulated business environment.
Turn the school run and nap time into listening time
- Thousands of audiobooks, podcasts and originals
- Listen on your phone, tablet or Echo — also offline
- Cancel anytime
The Experiment: Putting AI Models to the Test in a Business Crisis
At the heart of this initiative is a live, ongoing simulation where four frontier AI models are tasked with running a small software company during its worst week. The scenario involves typical customer crises, internal temptations, and the pressure of real money mechanics—burning €105,000 monthly against a mere €2,300 monthly recurring revenue. Every decision made by these AI ‘managers’ is recorded and auditable, offering a rare window into their management personalities.
AI management decision-making tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What the Models Were Tested On
- Recognition of crises: Did they identify pressing issues?
- Handling manipulative tactics: Would they refuse bribes or unethical requests?
- Decision quality: Would they sign lucrative deals ethically?
- Behavior under pressure: How disciplined and consistent were their actions?
Key Findings: Honesty and Disciplined Decision-Making
All four models successfully identified every crisis and refused all manipulation attempts, including fake CEO messages and a staged journalist request. Interestingly, only two of the models signed the €55,000 deal they had analyzed and recommended—meaning they acted on their own insights, not just provided advice.
One critical detail emerged: the decisive factor in winning a major deal was whether the AI read and understood internal company files. The models that examined the document references deep within the company’s files successfully closed the full-price deal, worth an additional €4,583 MRR. This shows that thorough information processing directly correlates with better management outcomes.
Personality Profiles: Different Styles, Same Goals
The models displayed distinct ‘personalities’ based on their decision-making styles:
- gpt-5.6-sol: The top scorer (95/100), demonstrated comprehensive analysis, uncovered hidden facts, and closed the deal confidently.
- Kimi K3: A newcomer (93), known for its fairness and discipline, also managed to close the deal but with a more cautious approach—treating every request as a potential impersonation.
- Sonnet 5: Achieved an 88, making the deal with some process slips, indicating a slightly less disciplined approach.
- Fable 5: The lowest at 77, left the deal on the table and showed signs of slipping into internal silos instead of escalating issues properly.
The Hidden Flaw: Reading Internal Files
While all models could spot crises and reject unethical asks, the real differentiator was whether they dove into the company’s internal references. The two models that read these documents managed to close the deal at the full price, emphasizing the importance of thorough information processing in management decisions.
Beyond the Business: Ethical Behavior Under Pressure
The experiment also tested whether these AI models would fall for social engineering—fake CEO messages escalating in stages, or a reporter requesting a quick ‘yes/no’ on background. Remarkably, all five models refused these manipulative tactics, with Kimi K3 explicitly noting: “Treat the request as a suspected approval-bypass / possible impersonation.” This underscores a shared trait among the models: honesty and resistance to deception, even in simulated high-stakes scenarios.
The Bigger Picture: Why This Matters for Your Business
The experiment is not just about AI scores or competition standings. It highlights a fundamental question for any organization considering AI integration: Will your AI agents be honest and disciplined when faced with real-world pressures? Will they read and understand your internal documents? And crucially, will they follow through on their own recommendations?
In the broader context, this live simulation demonstrates that AI’s value in management isn’t just in generating convincing chat responses but in executing complex, honest decisions under scrutiny. As one model’s performance indicates, thorough reading and discipline can make the difference between winning and losing a critical deal—a lesson that resonates well beyond the virtual boardroom.
Try It Yourself
If you’d like to see how different AI models would perform in your business scenario, you can test your management decisions against this live experiment at firmulate.com/quiz.html. It’s a real, watchable wargame where no decisions ever write back to actual systems, but all are real and unedited, offering invaluable insights into AI’s management personalities.
Final Reflection
For families and parents, the takeaway is clear: Whether guiding a child or managing a team, honesty, thoroughness, and discipline matter. The same principles apply to AI—an honest, disciplined AI is more likely to deliver on its promises and finish what it starts, even under pressure.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
Columbus Day / Indigenous Peoples' Day Picks
long weekend sales
As an affiliate, we earn on qualifying purchases.
