Imagine trying to cook a complex recipe—sometimes your ingredients align perfectly, and other times a crucial step gets overlooked. Now, picture AI models as the chefs, each with their own approach to managing a tough week in a bustling software company. Which AI chef would follow the recipe to the end, and which might leave the dish half-prepared? This isn’t just a culinary metaphor; it’s a real-world experiment revealing how different AI models handle management decisions under pressure.
Get kitchen staples and gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
The Experimental Kitchen: Putting AI Models to the Test
At Firmulate, a company dedicated to stress-testing AI decision-making, four frontier AI models were tasked with managing a small software firm through its worst week. The setup was rigorous: same customers, same crises, same temptations to cut corners. Every decision was recorded, versioned, and auditable, creating a transparent window into how each AI approached the challenges.
How Did They Perform?
- All four models identified every crisis, demonstrating a keen awareness of the mounting issues.
- All models refused manipulation attempts, maintaining integrity under social engineering tests like fake CEO messages and reporter tricks.
- Only two models signed the €55,000 deal their own analysis had earned, completing the core goal of the exercise.
Interestingly, the secret to winning the deal lay not in the immediate crisis responses but two documents buried deep within the company’s own files. Models that managed to read these references fully and incorporate them into their analysis secured the full-price deal, worth over €4,583 monthly recurring revenue (MRR).
Products Worth Considering
Traeger Woodridge Pro Wood Pellet Grill and Smoker, 970 sq. in.
Traeger Pro 22 Wood Pellet Grill and Smoker, 572 sq. in.
Le Creuset Enameled Cast Iron Round Wide Dutch Oven, 3.5 qt, Sea Salt
As an affiliate, we earn on qualifying purchases.
Decoding the AI Personalities
Among the models, the most thorough was Opus 4.8. It analyzed over 80 learned rules and delivered the deepest insights. Despite this, it finished last in closing the deal, because it left some opportunities on the table and slipped into internal processes instead of escalating issues. This highlights a critical insight: being thorough doesn’t always guarantee a successful outcome if discipline lapses under pressure.
The other models demonstrated different management styles:
- The Kimi K3 model, a newcomer, showed the cleanest discipline and successfully closed the deal.
- Sonnet 5 also closed the deal but with slightly more process slips, indicating a moderate management approach.
What This Means for Business and AI Integration
These experiments matter because they reveal that AI’s ability to follow through is more than just writing convincingly—it’s about integrity, attention to detail, and discipline when stakes are high. In a live company managing real money, these traits are critical. An AI that recognizes every crisis, refuses manipulation, and reads critical files deeply can be the difference between sealing a lucrative deal and losing it.
Real Money, Real Risks
The live company managed by these models has 13 synthetic employees, handling real money mechanics, burning €105k each month against a revenue of just €2.3k MRR. Every weekday, its decision process is watched and recorded, providing a rare glimpse into AI management in action.
The Bottom Line: Trust, Discipline, and the Cost of Work
The experiment underscores a key fact: in high-pressure management, the ability to stay honest and thorough is paramount. While all models spotted crises and refused manipulations, only some closed the deals effectively. This suggests that when deploying AI for management tasks, one must consider not just decision accuracy but also the model’s discipline and focus under stress.
For enterprises considering AI management solutions, the takeaway is clear: test your AI workforce before hiring it into your core operations. Firmulate’s live wargame lets you see how your chosen AI would perform in real-world crises—without risking your actual business. Visit firmulate.com/quiz.html to challenge your AI and see which personality shines.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
