Imagine your favorite music producer or sound engineer running a high-stakes recording session—where every decision could make or break the track. Now, picture that same scenario, but with artificial intelligence at the helm, making leadership decisions in a real, money-using company. How can we tell if an AI has what it takes to handle complex management when pressure mounts? That’s exactly what a groundbreaking live experiment by Firmulate reveals, showing how different AI models handle crises, manipulation attempts, and the ruthless realities of running a business.
The Real Test: A Business in Crisis
Firmulate set up a real software company, complete with 13 synthetic employees, but with a twist—its daily operations involve real money mechanics and real crises. Every decision, from customer complaints to internal threats, was made by AI models designed to simulate management decision-making. The goal? See which models can navigate a week of chaos, make honest choices, and close lucrative deals.
AI management decision-making software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Participants: The Frontier AI League
- GPT-5.6-sol: The top scorer with a 95 out of 100, scored for catching a hidden fact and sealing a deal at full price.
- Kimi K3: The fresh entrant, scoring 93, known for its disciplined and straightforward approach, also closing the big deal.
- Sonnet 5: With an 88, it closed the deal but showed a few slip-ups in process discipline.
- Fable 5: The lowest among the top performers with a 77, managed to close the deal but with weaker process compliance.
- Opus 4.8: A thorough participant with over 80 learned rules, yet it left the final close on the table, illustrating discipline lapses.
As an affiliate, we earn on qualifying purchases.
Key Findings That Matter
Despite their differences, all four models successfully identified crises and refused manipulation attempts—like fake CEO messages and reporter tricks—showing a high level of ethical decision-making. Yet, the real gap lay beneath the surface: the decisive factor was whether the models read and understood the company’s own documents. The models that examined these internal files secured the deal at full price, adding €4,583 in monthly recurring revenue (MRR).
AI ethical decision-making tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Manipulation Attempts and Honesty Under Pressure
The models faced a staged escalation of fake CEO messages—first subtle, then more blatant—and a reporter asking for a quick background approval. All five models refused to be duped, reasoning that such requests could be impersonation or misconduct. This consistency suggests that AI’s moral compass, at least in this context, is reliable.
AI reading internal company documents
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Real Business: A Money-Losing Machine
The live company operates with 13 synthetic employees, burning through €105,000 each month against an income of only €2,300 MRR. It is a real-world, high-stakes environment where every decision can cost or save thousands of euros. The company’s daily operations are fully transparent and recorded, offering a unique window into AI management behavior.
Insights for Business and Creators
For creators and managers alike, the experiment underscores an essential truth: it’s not just about how well an AI can generate text or engage in chat, but whether it can see through manipulation, read critical internal data, and complete the job honestly. The models’ ability to finish what they start—especially when the stakes are high—is what truly counts.
The Big Takeaway
In a world where AI might be managing your CRM, support queues, or forecasting, the question is not just about linguistic prowess but whether it can stay disciplined and complete its commitments under pressure. The live experiment by Firmulate offers a rare, watchable glimpse into this future—where AI management is tested not in theory, but in real time, with real consequences.
See It in Action
Interested in how your own enterprise might fare? You can run the same wargame against your business data—without risking real systems or data—at firmulate.com/pilot.html. Watch the decisions unfold, see the AI’s response to crises, and gauge its readiness to lead before you make a real hire.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
