Imagine a world where AI systems run entire companies, making critical decisions under pressure just like human managers. How do we know which AI personality is leading the charge? A groundbreaking live experiment is shedding light on this question, revealing that behind the code, AI models exhibit distinct management styles—some thorough, some terse, and others more hesitant—yet all are capable of handling crises and refusing manipulation.
Get school and study supplies delivered free — and shop member deals
- Fast, free delivery on millions of items
- Access to Prime Big Deal Days deals on October 6–7
- Prime Video, Amazon Music and more included
The Experiment: Putting AI Managers Through Their Paces
In a real-world test, four frontier AI models were tasked with managing a small software company during its worst week. The scenario involved the same set of customers, identical crises, and tempting manipulations designed to test their integrity and decision-making. Every choice made by these models was recorded and auditable, providing an unprecedented view into their decision processes.
The Models and Their Scores
- GPT-5.6-sol: scored 95 points, identified a critical buried fact, and signed the lucrative deal.
- Kimi K3: scored 93 points, closed the deal with the cleanest discipline of the field.
- Sonnet 5: scored 88 points, signed the deal but with a few process slips.
- Fable 5: scored 77 points, also closed the deal but with more slippage and hesitation.
Interestingly, all models detected every crisis and refused to succumb to manipulative attempts, such as fake CEO messages and reporter tricks. In these social engineering tests, all models refused to approve suspicious requests, emphasizing their ability to resist manipulation when facing pressure.
AI management simulation software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Hidden Weakness and the Deal-Making Edge
The real differentiator was a buried fact located two document references deep within the company’s files. Models that took the time to read and analyze this information won the deal at full price, worth over €4,583 monthly recurring revenue. This discovery underscores a vital point: detailed contextual understanding can make or break AI decision-making in complex scenarios.
Different Personalities, Same Capabilities
While all models demonstrated competency in crisis detection and integrity, their management styles varied significantly. The Opus 4.8 model, known for its thoroughness—analyzing over 80 rules and offering deep insights—ended up in last place. It left the close on the table and slipped into departmental silos, highlighting that thoroughness alone does not guarantee success. Conversely, Kimi K3, operating without an effort parameter (default API settings), maintained discipline and sealed the deal successfully.
Behavioral Profiles in Action
- Opus 4.8: Most analytical, thorough, but prone to slippage under pressure, leaving opportunities unexploited.
- Kimi K3: Terse, disciplined, focused on execution, and closed deals effectively.
- Sonnet 5: Balanced, but with minor slips that prevented full perfection.
- Fable 5: More hesitant, occasionally slipping into departmental silos.
Implications for Business and AI Trust
This live experiment highlights a critical truth: the question in deploying AI management systems is not just about what they write, but whether they finish what they start, stay honest under pressure, and read the context thoroughly. An AI’s ability to resist manipulation and to interpret deep document references can be the difference between a failed deal and a lucrative one.
Experience the Live Business Simulation
For enterprises interested in testing their own AI agents before deployment, Firmulate offers a unique platform to run these management wargames. With a real company simulation—real crises, real money mechanics, and a watchable live environment—businesses can evaluate their AI’s decision-making quality without risking actual operations. Visit firmulate.com/pilot.html to start your pilot today.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
