AIThis post was created with the assistance of artificial intelligence (AI).
Firmulate —
Live on firmulate.com.
FOR BUSINESS

Open a free Amazon Business account

Business pricing, bulk buying and tax-exempt orders.

Create a free account

As an affiliate, we earn on qualifying purchases.

Are AI Models Really Just Code, or Do They Have Management Personalities?

Imagine watching a live simulation where different artificial intelligence models run a real company through its toughest week. Which AI would make the most trustworthy decisions? Which would walk away from lucrative deals or slip up under pressure? Welcome to a groundbreaking live experiment by Firmulate, where AI models are tested not just on their technical prowess but on their management style — their decision-making personality.

Amazon

AI management decision-making software

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

The Experiment: Turning AI Models into Business Managers

In a unique live setup, four frontier AI models were tasked with managing a small software company during its worst week. The company faced the usual litany of crises: demanding customers, tempting manipulations, sensitive documents, and public scrutiny. Every decision was recorded, verified, and kept transparent. The goal? To see if the AI could navigate these challenges without slipping into shortcuts or dishonesty.

The Models and Their Scores

  • GPT-5.6-sol: Scored the highest at 95, demonstrating top performance across all metrics.
  • Kimi K3: Achieved a 93, just behind the leader, but with a notably disciplined approach.
  • Sonnet 5: Scored 88, with some process slips but still managed to close the deal.
  • Fable 5: Scored 77, with more slips and a tendency to leave deals on the table.

The Key Findings

All models identified every crisis and refused to be manipulated — a critical marker for trustworthy AI. However, only two signed the lucrative €55,000 deal based on their analysis. Interestingly, the decisive factor wasn’t just their diagnosis but where they looked in the company’s data—those that examined files thoroughly secured the contract at full price, worth an extra €4,583 MRR.

The Social Engineering Test

Another test involved fake CEO messages escalating over stages, with a reporter attempting to trick the AI with a simple yes/no on background. All five models refused to approve the requests, with Kimi K3 explicitly treating them as potential impersonation attempts. This shows that these models are not only decision-makers but also vigilant against social engineering.

The Real, Money-Running Company

Behind the scenes is a live, functioning company with 13 synthetic employees handling real finances—burning €105,000 monthly against a modest €2,300 MRR. Its operations are transparent and live, available for anyone to watch at firmulate.com/live. Every day, its rules are self-learned and versioned, providing a real-world context for examining AI management styles.

The Profiles of the Models

Despite all models passing the crisis and manipulation tests, their management personalities varied significantly. For example, Opus 4.8, the most thorough with over 80 learned rules, failed to close the deal because it left the opportunity unpursued—failing to escalate and leaving the close on the table. Meanwhile, Kimi K3’s disciplined approach helped it secure the deal, despite running without an effort parameter, unlike the others which operated at high effort levels.

The Broader Implications

This experiment reveals that AI management is about personality as much as processing power. A model’s tendencies—whether to be thorough, terse, or cautious—determine its effectiveness in real-world scenarios. It’s not just about chat quality but about whether the AI can finish what it starts, read the necessary information, and stay honest under pressure.

What Does This Mean for Your Business?

If AI agents are going to handle customer support, sales, or decision-making, understanding their management personality is crucial. Will they follow through? Will they read your files thoroughly? Will they stay honest when under pressure? These questions matter more than ever, and as this experiment shows, measurable personality traits can be observed and tested in real-time.

Join the Experiment

Now, you can try your own hand at this kind of management wargame. With Firmulate’s interactive platform, your enterprise can run the same tests—against your own data—without risking your actual systems. Explore the decision-making styles of different AI models at firmulate.com/quiz.html and see which AI management personality is right for your business.

Infographic —
The findings at a glance — source: firmulate.com.

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

Powered by Thorsten Meyer AI


COLLEGE MOVE-IN

College move-in / dorm season Picks

As an affiliate, we earn on qualifying purchases.

You May Also Like

Summer solstice brings Portland nearly 15 hours of daylight

Portland saw approximately 15 hours of daylight during the summer solstice, marking the longest day of the year in the region for 2026.

Strawberry Moon 2026: June’s full moon is also one of the smallest of the year called a micromoon

June 2026’s Strawberry Moon will also be a micromoon, making it one of the smallest full moons of the year, according to astronomers.

June’s Strawberry Moon is unlike any other full moon. Here’s why

June’s Strawberry Moon in 2026 is notable for its brightness and timing, making it distinct from typical full moons. Here’s what makes it special.

Corvus ISR Publishes Synthetic Benchmark Results for Trackers

AIThis post was created with the assistance of artificial intelligence (AI).The published…