Imagine running a real company through its worst week—crises, manipulations, and tough choices all at once. Now imagine testing different artificial intelligence models to see which ones can handle it best. This isn’t science fiction; it’s happening now, live, at firmulate.com. The results are surprising and reveal the true personality of these AI managers.
The Live AI Management Experiment: Putting AI to the Test
At the heart of this groundbreaking experiment are four frontier AI models tasked with managing a small, real-world software company. The company faces the same set of challenges—crises with customers, tempting offers, and complex decision-making—every day. But here’s the twist: each AI model runs the exact same scenario, and every decision is carefully recorded and made transparent.
The goal? To see which models can spot every crisis, refuse manipulative tactics, and ultimately close lucrative deals without bending the rules. The models include the most advanced versions available, like gpt-5.6-sol and Kimi K3, alongside others designed for thoroughness or speed. The results are as revealing as they are dramatic.
Key Findings: Competence, Integrity, and Decision Style
- All four models identified every crisis presented to them, showing remarkable situational awareness.
- Each model refused every attempt at social engineering, including escalating fake CEO messages and a staged reporter trick—showing they can resist manipulation.
- Despite similar diagnoses, only two models successfully signed a €55,000 deal that their own analyses identified as the best outcome. The other two either hesitated or left the deal on the table.
The decisive factor? The models that read and understood information buried two documents deep in the company’s files were able to make the full, accurate decision and close the deal at full price (+€4,583 MRR). Those that didn’t—like the Opus 4.8 model—missed this crucial detail and left money on the table.
AI management decision-making tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Personality of AI Managers: Terse, Thorough, or Honest?
Interestingly, each AI displayed a distinct management style during the week:
- Opus 4.8: The most thorough, with over 80 learned rules and deep analysis, yet the slowest to act decisively. It left the closing on the table and slipped in discipline, writing attempts into a locked department instead of escalating.
- Kimi K3: The newcomer scored just slightly below the top but ran without an effort parameter, making it more disciplined and straightforward in its dealings.
- Sonnet 5: Balanced performance, closing deals but with a few slips in process discipline.
- gpt-5.6-sol: The top scorer, with a perfect record for detecting untrustworthy requests and closing the deal at full price.
This diversity in style illustrates that AI models are not just calculators—they have personalities that influence their management decisions and their reliability under pressure.
As an affiliate, we earn on qualifying purchases.
Trust and Integrity in AI Decision-Making
The experiment also tested social engineering tactics, such as staged CEO messages and a background enquiry from a reporter. All five models refused to cooperate, citing suspicion or possible impersonation, with Kimi K3 explicitly treating the request as a suspected approval bypass.
This ability to resist manipulation is crucial. As AI models increasingly touch sensitive parts of business—customer support, CRM, forecasting—their integrity becomes paramount. The experiment shows that AI can be trusted to refuse undue influence, provided it’s trained and tested appropriately.
AI deal-closing automation tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Why This Matters for Your Business
Most companies focus on AI chat quality or superficial performance metrics. But this experiment shifts the focus to management qualities: can an AI finish what it starts, understand the context deeply, and stay honest under pressure? These are the traits that determine whether an AI can truly support or even replace human decision-makers.
Many of these models are already running live at companies — managing workflows, making decisions, and handling crises. The question is: are they doing so effectively and ethically? The experiment at firmulate.com/quiz.html offers a simple, engaging way to test your understanding of these AI personalities and their management styles.
AI integrity and security solutions
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
See It Live and Decide for Yourself
The real company involved in this experiment operates every business day with 13 synthetic employees and real money mechanics—burning €105k a month against €2.3k MRR. Its decision-making is transparent and continuously monitored, giving you a rare glimpse into how AI models manage real economic risks.
To explore these insights further or to run your own AI management wargame, visit firmulate.com. Experience firsthand how different models perform under pressure, or test your own AI workforce before you hire it, ensuring it’s equipped to handle your business’s toughest moments.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
