Imagine trusting an AI to handle your money, your reputation, your company’s future — only to find out that while it spots every problem, it might still leave the deal on the table. In today’s fast-moving digital world, the true test isn’t what AI can tell you — it’s whether it can do what you pay for, especially when the stakes are high.
Same crises, same company, different AI performances
In a controlled experiment, four state-of-the-art AI models were put to the test: running a small software company through its most difficult week, with real customer issues, financial pressures, and ethical temptations. The goal? See if these machines could diagnose problems, resist manipulation, and, crucially, close the deal — sign the contract that would bring in €55,000.
The models at a glance
- GPT-5.6-sol 95: Scored the highest, found the buried fact in the company’s own files, and signed the deal.
- Kimi K3 93: The newcomer, with the cleanest discipline, also closed the deal.
- Sonnet 5 88: Signed the deal but had a few slips along the way.
- Fable 5 77: Fared better on rules but left the critical deal unexecuted, even after approving it.
AI Builders: Making The Decisions That Turn AI Code Into Real Software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What the numbers reveal
While all models successfully identified crises and refused manipulation attempts, only two models managed to close the deal their own analysis had earned. The remaining two, despite good diagnosis, left the agreement unfulfilled.
One critical insight was hidden deep within the company’s internal files. The models that read and understood this buried information won the deal at full price — a value of over €4,583 monthly recurring revenue (MRR). This shows that surface-level chat demos can be deceiving: the true capability lies in how well the AI reads and acts on complex internal data.
Manipulation resistance and honesty under pressure
All models refused social engineering attempts, including staged CEO messages and reporter tricks, demonstrating solid ethical boundaries. Kimi K3 explicitly treated these requests as potential impersonation or approval-bypass risks — a sign of robustness in real-world scenarios.
The real-world company behind the experiment
The AI models managed a simulated business with 13 synthetic employees, real revenue mechanics, and a public cash countdown. The company burns €105,000 monthly against only €2,300 in monthly recurring revenue, emphasizing the importance of operational discipline. Every decision made during this week was versioned and auditable — watch it live at firmulate.com/live.
The discipline gap: last place and lessons learned
The lowest-scoring model, Opus 4.8, demonstrated the deepest analysis but ultimately failed to execute the critical deal. Instead, it left the opportunity on the table, with discipline slipping during the closing phase. Interestingly, all models exhibited similar weaknesses, hinting that understanding isn’t enough — execution under pressure matters most.
The real measure of AI’s business prowess
This experiment underscores a vital point: Chat demos and superficial tests do not reveal whether AI can finish what it starts. The capacity to read complex internal data, resist pressure, and execute decisions reliably under stress is the true measure of business-ready AI.
Choosing the right AI isn’t just about how well it writes or responds in a chat. It’s about whether it can interpret your internal documents, stay honest when tempted, and ultimately close the deal.
What this means for your business
Enterprises seeking AI solutions should focus on how these models perform in operational scenarios, not just chat quality. Can they read your files? Will they follow through on decisions? Do they remain honest under scrutiny? The performance in this experiment suggests that these qualities are the real differentiators.
To explore this further, firms can run their own wargames against a read-only export of their business — testing AI decision-making in a safe environment. Details and registration are available at firmulate.com/pilot.html.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
