Imagine managing a small travel company where every decision matters, but instead of human staff, an AI team is in charge. Now imagine watching that AI navigate real crises, make tough calls, and even refuse manipulation attempts — all in real time. This is not a sci-fi scenario but the live experiment conducted by Firmulate, where an AI-driven company fights for survival while open to the world.
Get travel and outdoor gear delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
The Living Lab of AI Management
At the heart of this experiment is a small, fully operational software company run entirely by artificial intelligence models. Instead of human employees, there are 13 synthetic ’employees’ making decisions daily, from handling crises to negotiating deals. What makes this setup extraordinary is that it’s publicly accessible — you can watch the company’s daily struggles unfold at firmulate.com/live. Every workday is versioned, every decision is logged, and the company is in a constant state of trial and error.
Products Worth Considering
DJI Osmo Mobile 6 Smartphone Gimbal Stabilizer with Extension Rod, Slate Gray
Bose QuietComfort True Wireless Earbuds with Active Noise Cancellation, Black
Bose QuietComfort Wireless Bluetooth Headphones, Cypress Green
As an affiliate, we earn on qualifying purchases.
Real Money, Real Risks
Despite the AI’s impressive capabilities, the company is far from profitable. It burns €105,000 each month while generating only €2,300 in monthly recurring revenue. This stark imbalance underscores the high stakes of AI management — it’s not just about making decisions but making the right ones that lead to sustainable growth. The public cash countdown adds urgency: every decision could be the difference between survival and collapse.
The Experiment: Testing AI in a Crisis Simulator
Firmulate’s experiment put four leading AI models through the same grueling week of challenges. Each model faced identical crises, customer demands, and temptations to cheat — like manipulating data or bending rules. All models correctly identified every crisis and refused every manipulation. Yet, only two secured the €55,000 deal their own analysis had warranted. The other two models, despite similar diagnoses and pitches, failed to close the agreement.
The Hidden Weakness in Files
Interestingly, the decisive factor for closing or losing the deal was not in the immediate customer interactions but buried two documents deep within the company’s own files. The models that read these references had the edge and ultimately won the contract at full price, adding €4,583 to the company’s monthly recurring revenue. This highlights a crucial insight: in complex decision-making, the ability to dig deep into internal data can be the difference between success and failure.
Standing Up to Social Engineering
The AI models also faced social engineering tricks, including staged messages from a fake CEO escalating requests step-by-step. All five models refused to be manipulated, demonstrating resilience against impersonation and approval-bypass tactics. Kimi K3, one of the models, articulated its reasoning clearly: it regarded the request as a suspected impersonation, avoiding a risky shortcut.
The Reality of Building AI Leadership
This live company, with its manual, rule-based system of over 680 learned rules, shows that AI can mimic management decision-making under pressure. But it also reveals the current limits: the most thorough model, OPUS 4.8, performed the worst in terms of closing deals due to slips in discipline and escalation. The experiment demonstrates that technical thoroughness does not always translate into effective management, especially when discipline wanes.
Implications for Business and Travel
For travel and outdoor companies considering AI integration, the lessons are clear: success depends not just on AI’s ability to generate compelling chat but on its capacity for strategic, honest decision-making under stress. Will your AI help you close deals, read critical internal documents, and resist manipulation? Or will it falter when the pressure is on? The Firmulate experiment makes these questions tangible and immediate.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
