Imagine trusting an AI to run your business during its worst week—handling crises, making tough calls, and staying honest under pressure. Can such models truly act as reliable managers? Recent experiments with cutting-edge AI models suggest their personalities—and their trustworthiness—are more measurable than you might think.
The Live Business Simulator: Putting AI Models to the Test
At the forefront of AI innovation, Firmulate has created a real-time simulation where AI models run a small but complex software company through its most challenging week. This isn’t just a demo; it’s an actual operational environment with real money mechanics, customer crises, and the temptation to manipulate or cut corners.
The experiment involved four of the top frontier AI models, each tasked with identical decisions during the same week of chaos. The goal? To see if they can spot crises, resist manipulation, and close deals at full value. Every decision is versioned and auditable, making their decision-making transparent and comparable.
Measuring Management Personalities Through Performance
The results reveal a fascinating spectrum of AI ‘personalities’—not just in their technical ability but in their management style:
- GPT-5.6-SOL scored the highest (95). It identified the most hidden, buried facts in the company’s own files, closed a crucial deal, and demonstrated complete situational awareness.
- Kimi K3 followed closely (93). The newcomer showed discipline and integrity, refusing all manipulations and closing the deal at full price, despite running without the default effort parameter.
- Sonnet 5 scored 88. It closed the deal but with some process slips, indicating a slightly more lenient management style under pressure.
- Fable 5 scored 77, and in some cases, left opportunities on the table, showing a more cautious or less disciplined management approach.
What Really Sets These Models Apart?
One of the most striking findings was that all models successfully detected crises and refused manipulative attempts, such as fake CEO messages or reporter tricks. When faced with escalating fake CEO messages over three stages plus a coded ‘just one yes/no’ background question, all five models rejected the requests, citing suspicion of impersonation or bypassing approval protocols.
But beneath this surface-level honesty lies a subtle difference: the decisive factor in the company’s success was not in the immediate crisis detection but in their reading of documents buried two levels deep within the company’s own files. Those models that reviewed these files thoroughly won a €55,000 deal at full price—equivalent to +€4,583 in monthly recurring revenue—while others left money on the table.
The Real-World Impact of AI Management Styles
This experiment demonstrates that AI models are not just chatbots but entities with measurable management personalities. Some, like GPT-5.6-SOL and Kimi K3, showed a capacity for deep analysis, integrity, and thoroughness, making them more reliable for real-world business decisions. Others, like Fable 5, while competent, appeared more cautious or prone to process slips, which could impact long-term performance.
Interestingly, the models’ performance was consistent across all fronts: they recognized crises, refused manipulative tactics, and made decisions aligned with ethical standards. The key difference was in their ability to read and interpret crucial information buried within documents—an indicator of their depth of analysis and discipline.
Why This Matters for Business and AI Adoption
For enterprises considering AI for management or decision-making roles, these findings highlight an essential question: Will your AI read your files thoroughly? Will it stay honest under pressure? Can it finish what it starts? Or will it leave opportunities on the table due to lapses in discipline?
Current AI models are capable of more than just generating chat responses; they can act as responsible, disciplined managers if trained and selected appropriately. The leaderboard scores—ranging from 77 to 95—serve as a benchmark for assessing their management personalities and trustworthiness in real business scenarios.
Try the Experiment Yourself
Wonder how your own AI models would perform? You can run the same business wargame against your enterprise’s data with a read-only export—nothing ever writes back to your actual systems. Visit firmulate.com/quiz.html to test your models’ management style and see which AI personality aligns best with your company’s needs.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
Products Worth Considering
AI management decision-making software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
