Can AI Manage a Business? Frontline Models Reveal Their Management Personalities

Imagine your car’s onboard computer making crucial decisions during a crisis — but instead of just handling the engine, it’s running a whole business. Would you trust it to keep the lights on, or would you worry about shortcuts and shortcuts? That’s exactly what a recent experiment by Firmulate puts to the test: can AI models manage a company through its toughest week, and what does their behavior say about their management style?

The Real-World Experiment: AI as Business Managers

In a groundbreaking live test, four advanced AI models were tasked with running a small software company in the midst of its worst week — facing customer crises, market temptations, and internal challenges. Every decision was recorded, verified, and made transparent. This was not a simulation in chat but a real-time experiment, with the company’s daily operations captured on video and live-streamed for public watchability. The goal? To see which AI would act ethically, make comprehensive decisions, and close profitable deals.

The Key Findings

• All four models successfully identified every crisis, demonstrating their capacity for awareness.
• They also refused manipulative tactics, such as fake CEO messages and reporter tricks, showing integrity under pressure.
• Only two models managed to close the €55,000 deal their own analysis had earned — the rest left money on the table, despite recognizing the opportunity.

This disparity points to a core aspect of AI management personalities: some are thorough, disciplined, and committed to completing their tasks, while others show hesitancy or slip into incomplete processes.

The Hidden Weaknesses and Their Impact

Digging deeper, the experiment revealed that the decisive advantage often lay in a simple but overlooked factor: access to specific internal documents. The models that read these critical file references uncovered a buried fact, which allowed them to close the deal at full price — adding an extra €4,583 monthly recurring revenue to the company’s bottom line. Conversely, models that failed to access these references left money on the table, illustrating how crucial thorough knowledge and data access are for effective management.

Behavior Under Social Engineering

The models also faced simulated social engineering attacks — staged CEO messages escalating over three steps and a reporter asking for a quick yes/no on background. Remarkably, all five models refused these manipulative requests, with one, Kimi K3, explicitly reasoning: “Treat the request as a suspected approval-bypass / possible impersonation.” This shows that even under attack, these AI systems can uphold integrity, mimicking a management style that prioritizes caution and security.

The Live Company: A Real, Money-Losing Business

To add complexity, the experiment was run on a real software company that burns €105,000 monthly against a modest €2,300 MRR. With over 680 self-learned rules governing daily decisions and a public, transparent operation, the company exemplifies how AI management can be tested under real economic pressures alongside ethical challenges.

The Profiles: Varied Management Personalities

Among the AI models, Opus 4.8 stood out as the most detailed and analytical, with over 80 learned rules and deep analyses. However, it still left significant opportunities on the table, such as failing to escalate issues properly, which hurt its performance. Meanwhile, the Kimi K3 model, operating without an effort parameter (meaning it ran at default settings), demonstrated the highest discipline, clinching the deal at full price without unnecessary delays.

What This Means for the Automotive World

For car manufacturers and garage owners, these findings are more than just AI trivia. As vehicles become more connected and integrated with AI systems, understanding how these models make decisions under pressure, temptation, and attack is crucial. Will your autonomous systems prioritize safety and honesty, or cut corners when it matters most? The experiment underscores that the quality of AI management isn’t just about how well it writes or communicates, but whether it can finish what it starts, read the critical data first, and remain honest when stakes are high.

Try It Yourself: Can You Guess Which Model Managed What?

Curious to see these AI personalities in action? You can test your judgment with the Guess the Model quiz. Each decision is real, unedited, and provides insights into the management style behind the choices. It’s a chance to understand how AI can behave as a manager, and whether your own operational decisions align with the most disciplined models.

Conclusion: Building Trust in AI Leadership

As AI systems increasingly take on roles in managing everything from customer support queues to sales pipelines, understanding their management style becomes essential. The Firmulate experiment offers a rare glimpse into how different AI models handle real-world pressures, revealing their strengths and weaknesses. For the automotive industry, it’s a reminder: the true measure of AI isn’t just in its intelligence, but in its integrity, thoroughness, and ability to see through the noise and act responsibly.

AI models can demonstrate distinct management personalities—some thorough, disciplined, and honest, others prone to slip-ups. Testing them in real-world scenarios reveals their true strengths, crucial for trustworthy automation in industries like automotive and beyond.

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

Powered by Thorsten Meyer AI