Imagine a manager so honest that they refuse to bend rules, yet another that keeps their responses brief and to the point. Now, what if these behaviors are actually traits of different AI models? As businesses increasingly rely on artificial intelligence for decision-making, understanding these ‘personalities’ becomes crucial—not just for efficiency, but for trust and integrity in management.
Testing the Moral Compass of AI in Business
In a groundbreaking live experiment, four frontier AI models were put through the same intense week of managing a small software company facing real crises, customer demands, and tempting shortcuts. The goal was simple: Could these AI ‘managers’ recognize and handle ethical dilemmas while running a company that burns €105,000 monthly against a revenue of just €2,300?
What makes this experiment particularly revealing is that every decision made by the AI was fully recorded, versioned, and auditable. This means that not only can we see what decisions they took, but also understand how their underlying ‘personality’ influences their actions. The models ranged from highly thorough to more terse and disciplined, mimicking different management styles.
The Results Are Eye-Opening
- All four models detected every crisis and refused every attempt to manipulate them—be it through fake CEO messages or subtle pressure tactics.
- Only two of these models managed to sign the €55,000 deal that their own analysis justified. The other two, despite understanding the situation, left the deal on the table.
- Interestingly, the decisive edge came from reading a hidden document in the company’s files—something not apparent in immediate customer interactions. The models that accessed these references clinched the deal at full price, adding €4,583 in monthly recurring revenue (MRR).
AI management decision-making tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Personality of AI Managers Revealed
Among the contenders, the model called Opus 4.8 was the most thorough, analyzing over 80 learned rules and providing deep insights. Yet, it finished last in the final tally because it left opportunities unexplored—leaving a crucial deal unclosed and slipping into less disciplined behavior.
Conversely, Kimi K3, the newcomer, showed the most discipline. Operating without an effort parameter (meaning it ran at default ‘high’ effort), it consistently refused shady requests and maintained integrity, earning a high score of 93 out of 100. This model’s refusal to bend under social engineering tactics—fake CEO messages staged over multiple stages and a reporter trick—demonstrates its strong ethical stance.
What Does This Mean for Business and Psychology?
This experiment offers a glimpse into the emergent ‘personalities’ of AI models—traits that can either bolster or undermine a company’s integrity. For organizations integrating AI into decision-making, the key takeaway isn’t just whether the AI can write well or handle support tickets, but whether it can stay honest, read relevant documents thoroughly, and follow through on commitments.
Furthermore, these AI personalities could mirror human traits, providing insights into how different management styles might behave under stress—ruthless, thorough, terse, or disciplined. Recognizing these traits could help in designing AI managers aligned with corporate values and compliance standards.
Experience It Live
At firmulate.com/live, you can watch this experiment unfold in real time. The company runs every business day, with real money mechanics, 680+ self-learned rules, and transparent decision logs—making it a compelling testbed for AI management’s future.
If you’re a business leader or a psychologist interested in behavioral dynamics, you can even run your own ‘wargame’ against a read-only export of your company data. This allows you to see how different AI models might handle your unique challenges without risking real-world consequences. Learn more at firmulate.com/pilot.html.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
