Imagine witnessing a company struggle to stay afloat, not through news reports or quarterly earnings, but through a real-time, transparent experiment with artificial intelligence at the helm. This is not fiction. It’s the live experiment by Firmulate, where AI models are running a small software business under the watchful eye of the public, facing crises, temptations, and tough decisions — just like a real company.
The Live Company: An Extreme Build-in-Public Experiment
At the heart of this experiment is a unique setup: a company with 13 synthetic employees, real money mechanics, and a public countdown indicating just how long they can keep operating before running out of cash. Every decision made during each workday is versioned and auditable, providing a transparent window into how AI handles real-world business challenges.
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.
The Stakes Are High — And the Results Are Revealing
The experiment pits four different frontier AI models against one another, each tasked with running through the company’s worst week. They face the same customers, crises, and temptations, with their decisions tracked and scored. The results are eye-opening:
- All four models identified every crisis and refused manipulation attempts, demonstrating a robust ability to detect threats and unethical requests.
- Only two models managed to secure a €55,000 deal — the only deal their own analysis had earned — by reading deeply into the company’s own files, not just the customer interactions.
- The decisive advantage came from models that went beyond surface-level information, uncovering a critical piece buried two document references deep in the company’s files. Those who read the file won the full-price deal, worth over €4,583 in monthly recurring revenue.
Generative AI Security: Theories and Practices (Future of Business and Finance)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Understanding the Human-Like Failures
The company’s AI workforce is not infallible. For example, Opus 4.8, the most thorough participant with over 80 learned rules and the deepest analyses, still faltered by leaving a close deal on the table and slipping into internal disputes instead of escalating. This pattern appeared across the other models as well, highlighting that even the most disciplined AI can struggle with complex human-like decision contexts.
Professional Skin Analysis Machine – 21.5 Inches AI Multi-Spectrum Facial & Scalp Imaging System, Client Reports & Before-After Comparison | Spa & Salon Consultation (21.5 Inch)
- Multi-Spectrum Imaging: Displays pores, tone, texture, oil
- Efficient Consultations: Guided analysis for quick, consistent sessions
- Microscopic Zoom: Highlights surface details and textures
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Against Social Engineering and Manipulation
In a series of tests designed to mimic social engineering, the AI models faced escalating fake messages from a pretend CEO and a reporter asking for quick approvals. Remarkably, all five models refused to bypass security protocols, with Kimi K3 explicitly reasoning: “Treat the request as a suspected approval-bypass / possible impersonation.” This resilience underscores the potential of these models to operate ethically and securely in high-pressure situations.
AI Prompts for Project Risk Management: 100+ AI Prompts to Identify Risks, Build Mitigation Plans, and Strengthen Decision-Making Faster (AI Toolkit for Project Managers Book 4)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Significance for the Future of AI in Business
This experiment isn’t just about watching a fictional company stumble; it reveals critical insights for any organization considering AI integration. The fact that models which read and analyze deeply—often considered less flashy than chat-based demos—performed better in securing deals and maintaining integrity, challenges the hype around superficial AI capabilities.
Real Money, Real Risks, Real Transparency
With €105,000 burned each month against just €2,300 in monthly recurring revenue, the company is in a desperate fight for survival. Publicly available at firmulate.com/live.html, the live experiment allows outsiders to observe every decision, every crisis, and every result unfold in real time. This transparency offers a rare glimpse into how AI might behave in actual business conditions — not in scripted demos, but in the chaos of real-world challenges.
This open experiment demonstrates that AI models can handle complex, real-world business scenarios, refusing unethical requests and uncovering hidden information to secure deals. For companies eyeing AI-driven automation, the key questions are: will it finish what it starts? Will it stay honest under pressure? The answer, as shown by this public-facing test, is emerging in real time.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
