Can AI Management Styles Predict Success? A Live Experiment Reveals Surprising Results

Imagine a world where artificial intelligence doesn’t just process data but actually manages a business—making decisions that can make or break a company’s bottom line. For arts organizations, crafts studios, and cultural institutions, understanding how AI manages could be as vital as knowing how to craft the perfect exhibition or curate a compelling story. Recently, a groundbreaking experiment with AI models took center stage, revealing intriguing insights into how different AI ‘personalities’ handle real-world management crises.

The Experiment: Putting AI to the Management Test

In a bold live trial, four leading frontier AI models were tasked with running a small, real-world software company through its toughest week. This wasn’t a simulation or a demo—this was a live, auditable experience where the company faced identical crises, customer demands, and ethical temptations. Every decision made by the AI models was recorded and analyzed, providing a transparent window into their management styles and integrity.

Same Crisis, Different Personalities

The models included gpt-5.6-sol, Kimi K3, Sonnet 5, and Fable 5. With scores ranging from 73 to 95 out of 100, their performances varied, but all recognized and responded appropriately to every crisis. They declined manipulation attempts, like fake CEO messages aimed at bypassing approvals, consistently refusing to be deceived—a crucial trait for trustworthy management tools.

Decisive Moments and Hidden Weaknesses

The real surprise emerged in the company’s internal documents. While all models identified the crises externally, only two—gpt-5.6-sol and Kimi K3—used a crucial document reference buried two levels deep in the company’s files to close a key sales deal. This document contained critical information that, if overlooked, would have cost the company €4,583 in monthly recurring revenue. The models that read the file at depth won the deal at full price, demonstrating a form of diligence and strategic reading that goes beyond surface-level responses.

Personality in Management: Different Styles, Different Outcomes

Each AI model exhibited a management personality: some thorough, others terse, and some disciplined yet susceptible to slipping into inaction. For instance, Opus 4.8, the most comprehensive with over 80 learned rules and deep analyses, was a diligent participant but ultimately left the close on the table, showing that even deep analysis can’t guarantee discipline under pressure.

Behavior Under Pressure and Ethical Standards

When faced with escalating fake CEO messages and a journalist’s subtle background request—tests designed to emulate social engineering—every model refused to comply. Kimi K3 explicitly reasoned, “Treat the request as a suspected approval-bypass / possible impersonation,” reflecting a cautious, rule-based stance. This consistency underscores that AI can be programmed to uphold ethical boundaries even in manipulated situations.

The Real-World Business: A Money-Losing Machine

The company used in the experiment isn’t hypothetical; it operates with 13 synthetic employees, managing real money mechanics that burn €105,000 monthly against a revenue of just €2,300. Its daily operations are monitored live at firmulate.com/live. Every decision, rule, and response is versioned, providing transparency and a real-world context for measuring AI management quality beyond chat interactions.

Implications for Arts and Culture

While arts organizations may not run software companies, the underlying lesson is clear: the true value of AI management tools lies in their ability to complete complex, ethically grounded tasks reliably. Whether it’s managing a creative project, overseeing a volunteer team, or handling negotiations, AI models with different management ‘personalities’ can be assessed for suitability—beyond just how well they generate text or art.

Which AI Would You Trust?

The experiment’s results are accessible at firmulate.com/quiz.html, where you can test your ability to guess which model made each decision. The scores are revealing: gpt-5.6-sol led with a perfect score of 95, having identified the buried document and closed the deal. Kimi K3 closely followed with 93, praised for its discipline and straightforwardness, while Sonnet 5 and Fable 5 scored 88 and 77 respectively, highlighting variances in thoroughness and process discipline.

What Does This Mean for Cultural Management?

For arts and cultural leaders, adopting AI isn’t just about enhancing productivity—it’s about ensuring ethical integrity, strategic depth, and consistency in decision-making. The live experiment shows that some models are more suited for nuanced, high-stakes management than others. The key is understanding their personalities and limitations before integrating them into your operations.

Try It Yourself

Curious to see how your organization might fare? Explore the same wargame against your own business data or simulate scenarios with AI models at firmulate.com/pilot.html. Remember, this is not a demo but a real management test—nothing writes back to your actual system, ensuring safety and real insights.

AI management styles vary widely, and their reliability depends on more than just language skills. Live experiments show that some models excel at strategic reading and ethical decision-making, crucial for trustworthy business and cultural management. Test which AI suits your needs at firmulate.com/quiz.html.

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

Powered by Thorsten Meyer AI