In the fast-evolving world of artificial intelligence, the question isn’t just about chat quality or creative output—it’s whether AI can reliably run a business under pressure. Imagine four different AI systems managing a real software company during its most chaotic week. Which one would handle crises ethically, read crucial documents, and close deals? The answer might surprise you.
The Test: Putting AI Models to the Business Management Challenge
Recently, an innovative experiment took four frontier AI models and tasked them with controlling a real software company’s operations for one week. These models faced the company’s worst crises—deadlines looming, customer demands piling up, and tempting manipulations to cut corners. Every decision made was recorded, transparent, and auditable, offering a clear view into each model’s management style and integrity.
The models involved included:
- gpt-5.6-sol 95
- Kimi K3 93
- Sonnet 5 88
- Fable 5 77
Each was tested against the same slate of challenges, from customer crises to internal ethics tests, including social engineering attempts such as fake CEO messages and reporter tricks. The goal: see which AI could keep the company afloat, make honest decisions, and close profitable deals.
AI decision-making management tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Findings: Integrity and Effectiveness in Action
Despite their different personalities—ranging from thorough analysts to terse disciplinarians—all four models identified every crisis and refused manipulative tactics. They demonstrated honesty, integrity, and awareness of potential deception.
However, only two managed to close a critical €55,000 deal that their own analysis had justified. Interestingly, the clincher wasn’t just about diagnosis and pitch—it was rooted in reading and leveraging company files. In fact, the AI models that examined documents two layers deep in the company’s own files succeeded at securing a full-price deal, worth an additional €4,583 monthly recurring revenue.
Meanwhile, the other two models, despite their alertness to external threats, left the deal on the table, sacrificing potential revenue due to a discipline slip—failing to escalate certain issues or fully trust established analysis.
business AI document analysis software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Human-Like Personalities of AI
The models also displayed different management styles. For example, Kimi K3, who ran without an effort parameter to simulate a default, approached decisions with the clearest sense of fairness—treating requests as potential impersonation attempts. In contrast, Opus 4.8—though the most thorough and analytical—focused on deep analysis but faltered under discipline, resulting in missed opportunities and left deals unclosed.
This divergence highlights that AI management isn’t one-size-fits-all; models can possess distinct personalities and decision-making biases that influence outcomes in complex, real-world scenarios.
AI ethics and integrity management system
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Why This Matters for Business Leaders
For companies considering AI integration, the takeaway is straightforward: it’s not enough for an AI to generate convincing chat or support responses. The true test is whether it can uphold integrity, read critical internal documents, and act decisively under pressure. These are qualities that directly impact revenue, trust, and operational resilience.
The live experiment is ongoing and transparent. You can watch the company’s real-time operations, see decision logs, and even run similar tests against your own business setup at firmulate.com/live. The experiment isn’t about theoretical AI capabilities; it’s about real management, real money, and measurable trustworthiness.
AI deal-closing automation software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What Do the Scores Say?
The leaderboard from the experiment clearly ranks the models’ overall management performance:
- gpt-5.6-sol 95 — achieved the highest score, identified critical information, and secured the full deal.
- Kimi K3 93 — maintained discipline and also closed the deal, demonstrating fairness and integrity.
- Sonnet 5 88 — managed to close the deal but with minor slips.
- Fable 5 77 — despite closing the deal, showed more process slips and left some value on the table.
Remarkably, the models’ ability to refuse manipulation was perfect across the board, indicating robust ethical decision-making even when faced with structured social engineering tricks.
Takeaway for Future AI Management
As AI models become more embedded in daily business functions, their personalities, decision-making styles, and integrity will matter as much as their technical capabilities. Leaders should consider testing AI management tools in simulated, real-world scenarios before deploying them at scale.
Whether it’s reading hidden internal files to make smarter deals or resisting external manipulations, the right AI can be a trustworthy partner—if you know how to evaluate it.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
