Imagine your favorite fashion brand facing a crisis—tightening deadlines, tricky supplier negotiations, and the pressure to deliver flawless collections. Now, picture AI models managing that chaos, with some making smarter, more honest choices than seasoned managers. Curious? Welcome to the cutting edge of AI management, where the decision-making style of machines is about to get real.
The Experiment: Testing AI Management Personalities in a Real-World Scenario
In a groundbreaking live experiment, four leading AI models were tasked with managing a small software company through its most challenging week. This wasn’t a typical AI demo—each model faced the same set of crises: demanding customers, potential manipulations, internal miscommunications, and the temptation to cut corners for quick wins. The goal? See which AI could navigate these storms ethically, efficiently, and effectively.
The Setup and the Stakes
The company simulated is real, with real money mechanics: burning €105,000 monthly against a modest €2,300 monthly recurring revenue. Every decision was recorded, every crisis was identical across models, and all manipulations—like fake CEO messages or attempts to bypass approval—were tested. The models’ responses weren’t scripted or coached; they were autonomous and unfiltered.
The Key Findings: Honesty and Effectiveness Matter
- All four models identified every crisis and refused every manipulation attempt, demonstrating a baseline integrity.
- Only two of the models managed to close a €55,000 deal they had analyzed and recommended—equivalent to a significant revenue boost.
- The decisive advantage lay in how deeply the models read and understood the company’s own files. The winner, GPT-5.6-sol, uncovered a hidden document reference that contained critical information, enabling it to close the deal at full price—adding €4,583 MRR.
Personality Profiles of the Models
The models exhibited distinctly different management styles:
- GPT-5.6-sol: Deep, thorough, meticulous—found the buried fact, closed the deal, and demonstrated comprehensive performance.
- Kimi K3: The newcomer, quick and disciplined—closed the deal too, with the cleanest process and strict refusal to manipulate.
- Sonnet 5: Slightly more prone to process slips but still effective in closing deals.
- Fable 5: Tended to leave opportunities unexplored, with discipline slipping under pressure.
Why This Matters Beyond the Lab
This live experiment underscores a crucial insight: the true test of AI management isn’t just its ability to generate convincing conversations. It’s whether it can uphold integrity, read deeply into information, and make decisions that drive real value. This is especially relevant for industries like fashion, where supply chain decisions, ethical brand management, and crisis response are critical.
The Human-Like Decision-Making Styles of AI
What’s fascinating is how these models display measurable personalities—a meticulous analyst, a disciplined rookie, or a cautious operator. For example, K3 refused to attempt manipulations without hesitation, emphasizing its approach to fairness and honesty. Meanwhile, the more thorough GPT-5.6-sol went a step further, unearthing a hidden reference that made all the difference in the deal.
The Risk and Reward in AI-Managed Business
In the real-world company running this experiment, every decision impacts cash flow. Right now, the company is losing €105,000 monthly with minimal revenue. The ability of AI to read files deeply, stay honest, and close deals at full value could transform operational efficiency—saving money and building trust in the process.
Try It Yourself: Run Your Business Wargame
If you’re a business leader curious whether your AI solutions can handle complex management tasks, you can run a similar exercise. Using Firmulate’s platform, companies can simulate their own crises and see how different AI models perform—without risking real money or data. The process is transparent, versioned, and safe, giving you a clear picture of your AI’s true capabilities. Visit firmulate.com/quiz.html to explore the interactive quiz and see which model might best fit your needs.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
