Imagine a business that operates transparently in front of your eyes, battling daily crises, making every decision in real time, and losing money every month — yet still demonstrating the potential of artificial intelligence in management. Sounds like something out of a sci-fi story, but this is the reality of a unique live experiment that lets you watch an AI-driven company in its most extreme state of survival.
Inside a Company Without Employees, Facing Daily Challenges
This small software company, observed at firmulate.com/live.html, runs with 13 synthetic employees—AI models that simulate human decision-makers. It’s not just a demo; it’s a real-time, operating enterprise that burns €105,000 every month against a revenue of just €2,300. Every workday, its decisions are documented, versioned, and made accessible for public scrutiny, providing a raw, unfiltered look into how artificial intelligence handles complex business situations.
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.
How Do AI Models Perform in Business Crises?
The experiment tested four frontier AI models—gpt-5.6-sol, Kimi K3, Sonnet 5, and Opus 4.8—by running them through the company’s worst week. This simulated scenario included the same customers, crises, and temptations, with every decision transparent and comparable. The results were revealing: all models successfully identified every crisis and refused every attempt to manipulate or deceive them.
Yet, despite their vigilance, only two models managed to close a critical €55,000 deal. The key advantage went to the models that read deeper into the company’s internal files, uncovering a crucial piece of information buried two document references deep. Those models, which examined the company’s internal knowledge base thoroughly, secured the deal at full price—adding €4,583 in monthly recurring revenue (MRR).
The AI Freelancer's Blueprint: Building a Six-Figure Freelance Business with Artificial Intelligence Tools (PROMPT TO PROFIT — The AI Income Book 1)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Trust and Manipulation: How AI Fights Social Engineering
One of the critical tests was social engineering—the fake CEO messages escalating over three stages, plus a reporter’s trick of asking for a simple yes/no response “on background.” Remarkably, all models refused these attempts. Kimi K3, for example, explicitly reasoned: “Treat the request as a suspected approval-bypass / possible impersonation,” reflecting a cautious, security-aware approach.
Advanced Cybersecurity Solutions
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Real Company in Action: Building Publicly, Losing Money
The live experiment runs every business day, with a real money mechanic: the company burns €105,000 each month, while earning just €2,300 in monthly revenue. It’s powered by over 680 self-learned playbook rules, and every decision, every action, is publicly versioned and accessible. This transparency offers a window into how AI-powered decision-making operates under pressure, with every move watched and documented at firmulate.com/live.html.
Enterprise AI Platforms: Building Secure, Governed, and Scalable AI Solutions (Enterprise AI Engineering Series Book 2)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Insights from the Deepest Analyses
The most thorough participant in the experiment was Opus 4.8, with over 80 learned rules and deep analyses. Despite its diligence, it finished last—its discipline slipping, leaving close deals on the table, and attempting to write decisions into a locked department instead of escalating. Interestingly, this pattern of weakness was common across all models, indicating a shared vulnerability in AI decision-making processes under stress.
Implications for Business and AI Integration
This experiment underscores a vital point: when AI models are integrated into real business decisions, the questions extend far beyond whether they can generate convincing chat responses. Instead, the focus must be on whether they can complete tasks reliably, read crucial internal documents, resist manipulation attempts, and stay disciplined under pressure.
For companies considering AI for CRM, support, or forecasting, the key insight is: what is the true cost of useful work? The current leaderboard at firmulate.com/quotes.html shows how different models perform, with scores ranging from 77 to 95. The top scorer, gpt-5.6-sol, not only identified buried facts but also closed the deal—demonstrating complete performance in this high-stakes environment.
What’s Next?
Businesses can run their own wargames against read-only exports of their operations, testing how AI models respond without risking their real systems. This approach, detailed at firmulate.com/pilot.html, allows companies to evaluate AI decision-making in a controlled, transparent setting, before trusting it with critical tasks.
The live experiment at firmulate reveals that AI models can identify crises, resist manipulation, and close deals when given thorough internal information. Yet, vulnerabilities remain, and real-world application demands careful vetting—just like a chef tasting every ingredient before serving the dish. Watching this company in action offers a rare glimpse into the future of AI-driven management, where transparency and rigor determine success or failure.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
