Which AI Model Runs the Show? Inside a Live Business Wargame Revealing AI Personalities

Discover How Different AI Models Show Their Management Personalities — Live and Unfiltered

Imagine testing AI systems as if they were real managers running a company. How do they handle crises, temptations, or ethical dilemmas? And more importantly, what do their decisions reveal about their personalities? This isn’t a thought experiment — it’s a live, ongoing experiment with frontier AI models at Firmulate, where real business challenges meet cutting-edge AI technology.

The Live Experiment: A Business in Crisis

At the heart of this experiment is a small, real software company, operating with 13 synthetic employees and real money mechanics. This company is losing €105,000 each month against €2,300 MRR, caught in a public countdown, with every decision meticulously logged and versioned. It’s a tense, watchable environment where AI models act as management teams facing the company’s worst week — same crises, same customers, same temptations.

Four frontier AI models participated, each running the company through identical scenarios. The models read and analyze company files, respond to crises, and make decisions about sales, customer care, and ethics. Their goal? Keep the company afloat, make deals, and avoid risky shortcuts or manipulative tactics.

Results That Speak Volumes

• All four AI models identified every crisis and refused every manipulation attempt, showing a strong grasp of integrity and risk management.
• Only two managed to close a €55,000 deal their own analysis had earned — an impressive feat, as they diagnosed the situation correctly and pitched convincingly. Yet, neither signed the deal when it came to actually sealing it, leaving potential gains on the table.
• Interestingly, the decisive advantage came from reading critical information buried two documents deep within the company files. Models that uncovered this hidden fact won the full deal, adding over €4,583 monthly recurring revenue.

Personality Traits of AI Models

Beyond raw decision-making, the experiment reveals distinct ‘personalities’ among the models:

• GPT-5.6-sol scored the highest (95) and successfully closed the most deals, suggesting a thorough, detail-oriented approach.
• Kimi K3, with a score of 93, was the most disciplined, refusing manipulative requests and closing deals cleanly, even without default effort parameters.
• Sonnet 5 scored 88 and closed deals with a few more process slips, indicating a slightly more relaxed style.
• Fable 5 with 77, also closed deals but left some opportunities unseized, hinting at less discipline under pressure.

Handling Ethical & Social Engineering Challenges

All models refused escalating fake CEO messages and a report trick designed to test their susceptibility to social engineering. Kimi K3 explicitly reasoned: ‘Treat the request as a suspected approval-bypass / possible impersonation.’ This consistency underscores a shared integrity across the models, regardless of their personalities.

The Broader Significance: Trust, Consistency & Cost

In the real world, AI agents are headed into CRM, support, and forecasting roles. But the key question isn’t just whether they generate fluent language — it’s whether they stay honest, read critical information, complete tasks, and avoid shortcuts, especially under pressure. The experiment at Firmulate demonstrates that some models excel at these qualities, translating to better business outcomes and lower costs.

For example, the models that uncovered buried information and closed deals earned significantly more revenue, with the full potential of over €4,583 extra monthly recurring income. Meanwhile, models with weaker discipline or less thorough analyses left money on the table, illustrating the importance of selecting the right ‘personality’ for AI management tasks.

How to Use This Knowledge

Business leaders can now test their own AI models against realistic scenarios, using Firmulate’s live wargame platform. This allows enterprises to see how their AI workforce handles crises, manipulations, and ethical dilemmas before deploying them in the real world—reducing risk and ensuring trustworthy AI management.

Visit firmulate.com/quiz.html to try the interactive quiz, and see which model’s management style matches your company’s needs.

Because these experiments are fully transparent, decision-makers can understand the strengths and weaknesses of each AI personality, making informed choices about where and how to incorporate AI into their operations.

Future Implications

This live AI management experiment is just the beginning. As models get better and more nuanced, their ability to act ethically, read deeply, and make strategic decisions under pressure will be critical. Whether for customer support, sales, or strategic planning, understanding the management personalities of AI models will become an essential part of enterprise AI strategies.

Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html

Powered by Thorsten Meyer AI