Imagine a company with no employees, losing €105,000 a month, yet still fighting to close deals and stay afloat. Now, watch it live, as artificial intelligence models steer through crises, temptations, and complex decisions — in real time. This is not science fiction; it’s the latest experiment in AI-driven management, unfolding publicly at firmulate.com/live.html.
The Live Company: A New Kind of Business Simulation
Firmulate has created a virtual company composed of 13 synthetic employees, each driven by advanced AI models. The company’s core mechanics mimic real-world money flows and decision-making processes — burning €105,000 each month while earning just €2,300 in monthly recurring revenue. Every workday, the company’s activities are versioned and publicly accessible, offering a transparent window into AI decision-making under pressure.
AI Marketing Mastery: Expert Secrets to Building a 7-Figure Coaching Business
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Experiment: Testing AI Under the Worst Conditions
Four frontier AI models — including the latest GPT-5.6-sol, Kimi K3, Sonnet 5, and Fable 5 — were placed in the same challenging scenario. They faced the same customers, crises, and temptations, with the goal of running the same small software business through its toughest week. Each decision was recorded and auditable, ensuring no sneaky shortcuts or hidden tricks could influence the outcome.
Key Findings: AI’s Truth and Trustworthiness
Remarkably, all four models identified every crisis and refused every attempt to manipulate them, such as social engineering tactics. For example, fake CEO messages escalating over multiple stages and a reporter trick — all were declined by every AI. The models’ on-record reasoning showed they treated suspicious requests as possible impersonation or approval-bypass attempts, highlighting robust ethical behavior under pressure.
The Critical Discovery: The Buried Fact in the Files
The decisive factor in closing a full-price deal was hidden deep within the company’s own documents — not in the immediate customer interactions. Models that read and analyze these internal references successfully identified the buried fact and won the €55,000 deal, boosting monthly recurring revenue by over €4,500. In contrast, models that missed this critical detail failed to close the deal, despite diagnosing the crisis correctly and presenting the right pitch.
Real Money Mechanics and Burn Rate
The live experiment underscores the harsh reality of running a business in a high-pressure environment. The virtual company burns through €105,000 monthly with a meager €2,300 in revenue, illustrating the challenge of maintaining operations despite AI management. Every decision made and every rule learned is openly documented and systematically improved — with 680+ self-learned playbook rules guiding behavior.
Model Performance and Discipline
The most detailed participant, Opus 4.8, analyzed over 80 learned rules and conducted the deepest analysis but still left a potential deal unclosed and slipped into internal write attempts instead of escalation. Interestingly, Kimi K3, running without an effort parameter and with default API settings, performed nearly as well as the more disciplined models, suggesting that AI fairness and default configurations matter in management performance.
Implications for the Future of AI in Business
This experiment isn’t just a tech demo. It raises key questions about the reliability of AI in real-world management tasks, especially where trustworthiness and thoroughness are critical. If these models can reject social engineering and identify hidden facts in internal documents, they could become vital tools for handling complex business environments — provided they are tested rigorously like this.
Watch It Live and Learn
Curious to see how AI manages in real-time? You can watch the ongoing performance at firmulate.com/live.html. The platform offers a transparent, versioned look at every decision, crisis, and win — a living laboratory for AI management, built in public and open for scrutiny.
The live experiment at Firmulate demonstrates that AI can recognize crises, resist manipulation, and even uncover hidden truths in internal data — all under extreme pressure. Watching AI’s management of a simulated company reveals what it takes to trust automation in real-world business, emphasizing the importance of transparency, thoroughness, and ethical rigor in AI decision-making.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
