Imagine a world where AI isn’t just crunching numbers but actually making management decisions—handling crises, closing deals, and even navigating the tricky waters of corporate trust. Sounds like sci-fi? Not anymore. At Firmulate’s live experiment, real AI models are running a small software company through its toughest week — and you can watch it all unfold.
The Experiment: Putting AI to the Test in a Real Business Crisis
What happens when different advanced AI models are tasked with managing the same company during its worst week? At Firmulate, four frontier AI models are running a simulated but authentic business environment, each facing identical crises, customer issues, and temptations to cheat or cut corners. Every decision they make is recorded, transparent, and comparable — giving us a rare window into how these models behave under pressure.
How the Models Perform
- All four models identified every crisis, from customer complaints to internal conflicts.
- Every AI refused manipulation attempts, such as fake CEO messages or reporter tricks, showing integrity in decision-making.
- Only two models managed to close a significant €55,000 deal earned through their own analysis — the rest left it on the table despite their diagnosis and pitch being identical.
Interestingly, the critical advantage for winning the deal was rooted not in superficial chat or surface-level responses but in the models’ ability to read and interpret a complex document stored deep within the company’s files. Those who read the full references secured the deal at full price, adding €4,583 MRR (monthly recurring revenue) to the simulated business.
AI management decision-making software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Personality Profiles of the AI Models
The models demonstrate measurable management personalities, much like human managers:
- gpt-5.6-sol: The top performer, with a score of 95, it uncovered hidden information and closed the deal, displaying thoroughness and integrity.
- Kimi K3: The newcomer scored 93, distinguished by the cleanest discipline — it refused to sign deals without full validation, emphasizing fairness and caution.
- Sonnet 5: Scoring 88, it managed to close the deal but with some process slips, indicating a slightly more relaxed approach to discipline.
- Opus 4.8: The most thorough in analysis, yet it scored 73 and left the deal on the table, showing that even deep analysis doesn’t guarantee closing under pressure.
These personalities reflect different management styles—some meticulous, others more cautious—and their decision-making can be observed and measured objectively.
As an affiliate, we earn on qualifying purchases.
Handling Ethical Dilemmas and Pressure
The experiment also tested how the models respond to social engineering tricks, such as staged CEO messages and a fake reporter asking for quick approvals. Remarkably, all models refused to be manipulated, with Kimi K3 explicitly treating suspicious requests as potential impersonation.
This ability to maintain integrity under pressure is crucial for real-world AI applications, especially in sensitive areas like customer relations, compliance, and corporate governance.
AI ethical decision support systems
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Live Business: An Ongoing Watchable Company
The experiment is not just a demo; it’s a live, functioning software company with 13 synthetic employees, real financial mechanics, and a public cash countdown. Every workday, the AI models operate within a complex environment worth over €4,583 MRR, burning €105k monthly, yet generating only €2.3k in revenue — a real test of decision quality and discipline.
The company’s operations are fully versioned and transparent, and anyone interested can observe at firmulate.com/live. It’s an open lab where you can see how AI managers handle crises, make negotiations, and stay honest — in real time.
AI negotiation and crisis management tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What Does This Mean for Businesses?
This experiment is about more than just AI tech; it’s about trust, discipline, and the ability to deliver measurable value. As AI begins to touch CRM systems, customer support, or forecasting, the critical questions aren’t about how well it can chat but whether it can see the whole picture, stay honest, and finish what it starts.
For enterprises thinking about integrating AI into management roles, Firmulate offers a unique way to test and compare their AI options before fully committing — all without risking real-world money or reputation.
Final Takeaway
In a landscape where AI’s decision-making prowess is often hidden behind impressive demos, this live experiment exposes the real personality and integrity of each model. It’s a compelling glimpse into a future where AI managers might not just process data but lead with discipline, honesty, and strategic insight—perhaps even better than some humans.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
