Imagine a real company, facing its worst week — customers demanding support, crises erupting, and tough decisions needing to be made under pressure. Now picture multiple AI models running this company, each with its own personality and approach, competing to see which can handle the chaos best. Welcome to the live experiment by Firmulate, where AI models are put through a management simulation that reveals their true personalities — and their potential.
The Live Test of AI Decision-Making
In a groundbreaking experiment, four frontier AI models were tasked with running a small software company during its most turbulent week. Each model faced the same set of crises, customer complaints, and temptations — from regulatory breaches to negotiation traps. The goal was simple: see which AI could best navigate the storm, make honest decisions, and successfully close a lucrative deal.
How the Experiment Worked
The models, including the recent GPT-5.6-sol, Kimi K3, Sonnet 5, and Opus 4.8, were all given the same challenges in a controlled environment. Every decision was timestamped and auditable, ensuring transparency. The company they ran had real money mechanics — burning €105,000 monthly against a small €2,300 monthly recurring revenue, with a live public dashboard for observers. This setup was designed to mimic real-world pressures, including customer demands, internal conflicts, and even social engineering attempts like fake CEO messages and reporter tricks.
Key Findings: Honesty and Competence
All four models identified every crisis and refused manipulation attempts. When it came to closing deals, only two models signed the agreement worth €55,000 that their analysis had earned. Interestingly, they all diagnosed the issues correctly and delivered the same pitch, but only two followed through and closed the deal. The others left money on the table, demonstrating a variance in discipline and decisive action.
The Hidden Weakness
The decisive factor wasn’t in the customer interactions but buried several layers deep in the company’s files. The models that read and understood these internal documents secured the full deal at an additional €4,583 monthly recurring revenue. This illustrates an essential insight: AI’s ability to access and process critical internal data can be the key to maximizing outcomes.
Social Engineering and Trust
The experiment also tested models against social engineering. Fake CEO messages, staged in three escalating stages, and a reporter trick asking for a quick yes/no on background — all five models refused to be manipulated. Kimi K3’s reasoning was clear: “Treat the request as a suspected approval-bypass / possible impersonation.” This demonstrates that these models can be programmed to recognize and resist typical manipulation tactics, crucial for real-world applications.
AI decision-making management tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Personality Profiles of AI Managers
Beyond raw performance scores, the experiment highlighted the distinct personalities of each model:
- gpt-5.6-sol 95: The star performer, who uncovered the hidden document, closed the deal, and demonstrated thoroughness.
- Kimi K3 93: The newcomer, running without an effort parameter and showing disciplined decision-making, also closed the deal at full price.
- Sonnet 5 88: Capable but with minor slips, closing the deal with slight process lapses.
- Opus 4.8 77: Deeply analytical, but more cautious, leaving money on the table due to over-discipline and avoidance of escalation.
These profiles suggest that AI management personalities are measurable and vary significantly, impacting their ability to close deals, handle crises, and maintain honesty under pressure.
As an affiliate, we earn on qualifying purchases.
Why This Matters for Your Business
As AI begins to interact more with your customer relationship management, support queues, or forecasting, understanding their management personality is crucial. The question isn’t just whether they can write well or simulate conversation; it’s whether they can finish what they start, read critical internal information, stay honest under pressure, and ultimately, deliver real value.
Try It Yourself
Interested in testing your own AI’s management style? You can run the same wargame against a read-only export of your business data at firmulate.com/quiz.html. This interactive quiz allows you to gauge how different AI models might perform in your unique environment, all without risking your actual systems.
AI cybersecurity social engineering resistance
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What’s Next?
Firmulate’s live site offers a window into the future of AI-driven management. The experiment is ongoing, with every workday recording new decisions, learning from failures, and sharpening each model’s profile. This isn’t just a test — it’s a demonstration that AI personalities matter, and their ability to act ethically and decisively in real-world pressures is measurable and critical.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
As an affiliate, we earn on qualifying purchases.
