Imagine managing a busy garage or auto shop where every decision counts—what if your AI assistant was tested not just on chat skills, but on how well it handles real crises, trusts your files, and stays honest under pressure? Welcome to a groundbreaking experiment that puts AI models through their paces in a live business simulation, exposing their personalities and integrity in real time.
Putting AI Management to the Test
At Firmulate, a pioneering company in AI-based management simulations, researchers have created a unique live experiment that simulates the worst week of a small software company. The twist? They ran four leading frontier AI models—each with different personalities—through the same scenario, with identical crises, customer demands, and temptations to cheat. The goal was simple: see which models could maintain integrity, make sound decisions, and ultimately close profitable deals.
AI management decision-making tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
How the Experiment Worked
The models faced a series of management challenges, including:
- Identifying key information buried in the company’s internal files—sometimes two documents deep
- Responding to manipulative tactics like fake CEO messages and media inquiries
- Deciding whether to sign a lucrative €55,000 deal based on their analysis
All decisions were carefully versioned and auditable, providing transparency into each AI’s reasoning process. Interestingly, while all models spotted every crisis and refused manipulative attempts, only two managed to close the deal and sign the contract, earning their own analysis.
The Surprising Findings
The most notable insight was that the decisive factor in sealing the deal was whether a model read and understood critical information stored within the company’s own files. The models that delved deep into these internal documents secured the full €4,583 MRR bonus for their company, demonstrating the importance of thorough analysis and trustworthiness.
For example, the model GPT-5.6-SOL scored highest with a 95 out of 100 and successfully closed the deal by uncovering the buried fact. Meanwhile, Kimi K3, a newcomer to the field, scored a 93 and also signed the deal—showing it was the most disciplined in refusing manipulation attempts, including fake CEO messages and media tricks.
However, not all models performed equally. Opus 4.8, which conducted a deeper analysis with over 80 learned rules, finished last in the rankings with a score of 73. Its discipline slipped, and it left money on the table by hesitating to escalate issues properly. Interestingly, all models demonstrated a shared ability to spot crises and refuse manipulative tactics, highlighting that honesty under pressure was a common trait across the board.
What This Means for Business and Automotive Management
For automotive service providers, this experiment underscores a vital point: the true value of AI in management isn’t just in generating convincing chat responses. It lies in whether the AI can finish what it starts, read your internal files carefully, and stay honest when under pressure. In a garage or service shop, this could translate into AI systems that better manage customer requests, read critical service histories, and resist external manipulation—leading to more reliable and profitable operations.
The Personality of AI Managers
One of the most fascinating aspects of this experiment is that these models showcase measurable management personalities. Some are thorough and disciplined, like Kimi K3, which prioritized integrity and completed its tasks without shortcuts. Others, like Opus 4.8, are more analytical but may slip under stress, risking leaving money on the table. These characteristics are quantifiable, making it easier for enterprises to choose an AI that matches their management style and operational needs.
Share and Watch the Live Results
This isn’t just an experiment to read about; it’s a live, ongoing demonstration. You can watch the same real company—running every business day—navigate crises, make decisions, and learn from its AI workforce at firmulate.com. For businesses curious about how AI can act as a trustworthy manager or decision-maker, this platform offers a unique chance to see models in action, run tailored wargames against your own business data, and understand their decision-making personalities firsthand.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
