Imagine if your fitness coach could not only tell you what exercises to do but also decide how to motivate you based on their own personality. Now, scale that idea up to AI managing real companies—deciding how to handle crises, negotiate deals, or even stay honest under pressure. The question isn’t just whether AI can talk well; it’s whether it can finish what it starts under stress. That’s exactly what a groundbreaking live experiment by Firmulate reveals.
The Live AI Management Showdown
Recently, a real software company was put through a grueling week of crises, temptations, and tough management decisions. The twist? Each decision was guided by a different AI frontier model—identical situations, but each with its own management personality. The models are not just chatbots; they are complex decision-making systems that have been tested against real business challenges, with their choices recorded and analyzed.
Who Were the Competitors?
- gpt-5.6-sol: The top scorer with a 95 out of 100 score, known for thoroughness and finding hidden details.
- Kimi K3: A newcomer with a clean record (93), ran at default API settings, displaying discipline and honesty.
- Sonnet 5: An honorable performer (88), with some slips in process but still closing deals.
- Fable 5: The least reliable (77), often leaving opportunities on the table and slipping in discipline.
All models successfully identified crises and refused to fall for manipulative tricks like fake CEO messages or media tricks. But when it came to sealing the deal, only two models actually signed the €55,000 agreement — a crucial performance metric. The others either hesitated or left money on the table.
The Key to Success: Reading the Deep Files
Interestingly, the decisive factor wasn’t just surface-level decision-making but the models’ ability to access and interpret company files. The winning models looked two layers deep into internal documents—information that was hidden from the typical customer-facing view. Those who read the files won the deal at full price, adding over €4,500 to monthly recurring revenue (MRR). This underscores a vital point: AI’s management effectiveness hinges on thorough data reading, not just surface interactions.
How Do Models Handle Stress & Manipulation?
During the experiment, the models faced simulated social engineering attacks: staged messages from a fake CEO escalating in seriousness and even a media inquiry. Remarkably, all five models refused to participate in manipulation attempts, citing concerns over impersonation or bypassing approval processes. It’s a testament to their design to prioritize integrity, even under pressure.
AI decision-making software for business
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What Does This Mean for Business AI?
For companies contemplating AI-driven management tools, the experiment offers compelling insights. It’s not enough for AI to generate plausible deals or conversations; it must also demonstrate discipline, reading comprehension, and honesty — especially when stakes are high.
The live site, firmulate.com, showcases this ongoing experiment. You can watch the AI models in action, see real decisions unfold, and gauge the kind of management style each model exhibits. The experiment isn’t just about scores; it’s about understanding whether AI can truly manage complex, real-world business scenarios.
Beyond the Scores: Understanding AI Personalities
In this trial, the models displayed distinct personalities:
- gpt-5.6-sol: Deeply analytical, thorough, and prudent—scored highest for finding hidden info and closing full-price deals.
- Kimi K3: Disciplined, honest, and straightforward—ran without effort parameters and still managed to close deals.
- Sonnet 5: A balanced performer, with minor slips but still effective in sealing deals.
- Fable 5: Less disciplined, more prone to leaving opportunities untouched, illustrating that personality traits affect management outcomes.
The Big Takeaway for Business Leaders and AI Developers
This is more than a game. As AI models become more integrated into the fabric of operations—CRM, customer support, forecasting—the questions aren’t just about their intelligence or language skills. They are about integrity, thoroughness, and discipline. Can your AI finish what it starts? Does it read your files carefully? Will it stay honest when under pressure?
The experiment by Firmulate makes these questions tangible, showing that the management personality of an AI can have measurable impacts on business results. It’s a chance for enterprises to run their own wargames, simulate crises, and assess whether their AI workforce is ready for prime time.
AI management models vary greatly in discipline and thoroughness. The real test isn’t just talk — it’s whether they can finish deals, read deeply, and stay honest under pressure. Watch the live experiment and see which models are ready to manage your business.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
