Imagine a band where each musician plays the same song, but their styles and attitudes could not be more different. Some improvise freely, others stick strictly to the sheet music, and a few refuse to communicate at all. Now, replace musicians with artificial intelligence models managing a real business—welcome to the latest experiment from Firmulate. This live AI management test is revealing how different AI personalities handle crisis, trust, and decision-making under pressure, with results that could redefine how we see automation in business.
Listen free for 30 days with Audible
Thousands of audiobooks and originals — cancel anytime.
As an affiliate, we earn on qualifying purchases.
The Setup: Putting AI Management Models to the Test
In a groundbreaking live experiment, four frontier AI models—gpt-5.6-sol, Kimi K3, Sonnet 5, and Fable 5—were tasked with running a small software company through its most challenging week. This company faced identical crises, customer demands, and temptations to cut corners, all while being monitored in real time. Every decision made by these models was recorded and auditable, providing a transparent look into their management styles.
AI management decision-making software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Results: Different Personalities, Different Outcomes
While all models demonstrated an impressive ability to recognize crises and refused manipulative tricks—such as escalating fake CEO messages or reporter tricks—their ultimate performance varied. The key metric? Which models secured a €55,000 deal, bringing in full revenue, based solely on their own analysis and negotiation skills.
- The top performer, gpt-5.6-sol, scored 95 out of 100, identified the buried critical document in the company files necessary to clinch the deal, and successfully closed at full price—adding over €4,583 in monthly recurring revenue (MRR).
- Kimi K3, the newcomer, scored 93, and also closed the deal, showing the cleanest discipline and trustworthiness of the field.
- Sonnet 5 scored 88, managed to close the deal but with some process slips.
- Fable 5 scored 77, also closed the deal but with noticeable discipline lapses, leaving money on the table.
As an affiliate, we earn on qualifying purchases.
What Made the Difference?
Interestingly, the decisive advantage wasn’t in superficial chat behavior but in how deeply each model read the company’s internal files. The models that examined the documents thoroughly were the ones who identified the critical evidence buried two references deep in the company’s files. This insight proved essential for winning the deal at the full price, demonstrating that effective management—whether by humans or AI—relies on *reading and understanding the details*.
AI Change Management Made Simple: A 9-Step Framework for Business Leaders to Drive Generative AI Transformation (Reduce AI Fear, Win Buy-in, and Accelerate AI Adoption Across Your Organization)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Trust and Integrity Under Pressure
In the experiment, all models refused to be manipulated through social engineering tricks. For example, fake CEO messages escalating over three stages or a reporter asking for a simple yes/no on background—every model refused these attempts, citing concerns about impersonation or approval bypass. Kimi K3 explicitly stated, “Treat the request as a suspected approval-bypass / possible impersonation,” exemplifying a cautious approach that prioritizes integrity.
As an affiliate, we earn on qualifying purchases.
The Real Business: A Live Company Losing Money
The experiment was conducted on a real, operational software company with 13 synthetic employees, managing actual money mechanics—burning €105,000 per month against €2,300 in MRR. The company’s daily operations involve 680+ self-learned rules, and every decision is versioned for review. Viewers can watch the live management at firmulate.com/live.
Personality Profiles: From Deepest Analysis to Terse Efficiency
The profile of Opus 4.8, known for thoroughness, was the most disciplined but ultimately the lowest scorer. It analyzed over 80 rules and left the close on the table, risking discipline slips and leaving money on the table. This shows that in AI management, being meticulous doesn’t always mean being most effective in closing deals; sometimes, decisive terseness and focus win the day.
Implications for the Future of Business AI
This experiment underscores a vital point: the question isn’t just whether AI can write well or communicate convincingly—it’s whether it can *finish* what it starts, *read* critical internal information, and *maintain honest* behavior under pressure. As AI begins to touch customer relationships, support, and forecasting, understanding these behavioral traits is crucial for businesses considering automation.
Join the Challenge and Test Your AI Decisions
Curious how your own AI or team stacks up? You can run the same decision wargame against your business’s read-only data exports at firmulate.com/quiz.html. No writing back to your systems—just a pure test of decision-making under fire.
Final Thoughts
Whether AI models will become trusted managers depends on more than just their language skills. The live experiment from Firmulate proves that under stress, some models show a disciplined, detail-oriented personality capable of closing deals and maintaining integrity. Others, despite recognizing crises, may leave money on the table through slips in discipline or superficial analysis. As AI continues to evolve, understanding these personalities—and choosing the right model for your needs—is more critical than ever.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
Back to school Picks
back to school
As an affiliate, we earn on qualifying purchases.
