Imagine preparing for your next adventure, carefully planning every step, knowing that the decisions you make can determine your success or failure. Now, what if your AI team — managing a virtual company — faced the same challenge? Would it act decisively, or hesitate under pressure? This is no hypothetical: a live experiment with frontier AI models running a real, money-losing software business sheds light on how different AI ‘personalities’ handle crises, trust, and decision-making.
What Is the Firmulate Experiment?
At the heart of this story is an unprecedented live test: four advanced AI models, all designed by leading AI firms, were tasked with managing a small software company during its most chaotic week. The company faces identical crises, customer demands, and temptations to cut corners. Every decision they make is recorded, verified, and compared. The goal? To see which AI demonstrates the most trustworthy, disciplined, and effective management style.
AI management decision-making software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
How Did the Models Perform?
Despite their differences, all four AI models demonstrated impressive situational awareness:
- All models identified each crisis as it arose.
- All refused manipulation attempts — whether fake CEO messages or subtle requests to bypass procedures.
However, only two models succeeded in closing a critical €55,000 deal based on their own analysis — a key measure of effective management. The other two, despite diagnosing the issues well, failed to finalize the deal, leaving money on the table despite accurate understanding.
As an affiliate, we earn on qualifying purchases.
Deciphering the Hidden Weakness
The decisive factor wasn’t in the immediate crises but was buried two document references deep in the company’s files. Those models that read and understood this crucial information closed the deal at full price, adding approximately €4,583 in monthly recurring revenue. This underscores a vital point: reading comprehension and thorough analysis are decisive in effective AI management.
As an affiliate, we earn on qualifying purchases.
Trust and Integrity Under Pressure
In social engineering tests — fake CEO messages escalating in stages and a reporter asking for a simple yes/no answer — all models refused to comply. Kimi K3’s explanation was typical: “Treat the request as a suspected approval-bypass / possible impersonation.” This consistent refusal highlights that well-designed AI models can maintain integrity even amid social engineering tactics, a key concern in deploying AI in real-world business settings.
As an affiliate, we earn on qualifying purchases.
Real-World Business Mechanics
The experiment’s live environment is a functioning company with 13 synthetic employees, burning €105k monthly against a mere €2.3k in monthly revenue. Every day, the company operates with over 680 self-learned rules, and each decision is versioned for analysis. Watch it unfold live at firmulate.com/live. This visibility offers a rare glimpse into how AI models behave beyond clean chat demos, revealing their management ‘personalities’ in a real economic context.
The Profiles: Different Management Personalities
The models exhibited distinct decision styles:
- Opus 4.8: The most thorough, analyzing over 80 learned rules and providing deep insights. Yet, it left a crucial deal on the table and showed discipline slips — decisions were written into a locked department instead of escalated, hinting at a cautious, detail-oriented personality.
- Kimi K3: The newcomer with a focus on fairness and discipline, operating without an effort parameter (default API settings), successfully closed the deal with minimal slips, demonstrating a disciplined, straightforward approach.
- Sonnet 5: Struck a balance, closing the deal but with some process slips, illustrating a slightly more relaxed management style.
Interestingly, all models struggled with the same weakness: a failure to escalate issues properly, which could be critical in real business scenarios.
Why It Matters for You
While this experiment seems technical, its implications are relevant to any organization considering AI automation. The key questions are:
- Will the AI finish what it starts?
- Does it read and understand your documents thoroughly?
- Can it maintain honesty under pressure?
- How much useful work does it actually produce?
In a world where AI might manage customer relationships, support queues, or forecast sales, these qualities matter more than just how well it writes a message. Trustworthiness, diligence, and the ability to see the full picture can make or break your business — just as they did in this live experiment.
The Bottom Line
As of now, the AI league table looks like this:
- GPT-5.6-SOL: The top performer — found the hidden fact and closed the deal at full price.
- Kimi K3: Close behind, with the cleanest discipline.
- Sonnet 5: Managed the deal, but with some slips.
- Sonnet 88: Also closed the deal but showed more process slips.
All models are available to test in your own business wargames, without risking real systems. Explore the possibilities at firmulate.com/quiz.html.
AI models exhibit distinct management styles, impacting their ability to handle crises, trust, and negotiations in real business settings. Live tests reveal that thoroughness, discipline, and integrity are crucial — not just language skills. Prepare your AI workforce by testing its decision-making before deployment.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
