Imagine a business that operates publicly, with no human employees, losing thousands of euros every month, yet still fighting to stay afloat. This isn’t fiction; it’s a live experiment revealing how AI models make critical management decisions under pressure. For those interested in personal finance and investing, it raises urgent questions about trusting AI to handle real-world operations — and how close we are to companies that run themselves, for better or worse.
The Live Experiment: An AI Company in Real Time
At the heart of this story is a company called Firmulate, an unusual enterprise that exists entirely in the open. It features 13 synthetic employees, each driven by different AI models, tasked with running a small software business through its worst week. Every decision, crisis response, and negotiation is tracked, versioned, and publicly accessible at firmulate.com/live.html.
This isn’t just a demonstration of AI chat capabilities. It’s a rigorous, real-world test of management decision-making under pressure, with tangible financial stakes. The company burns through €105,000 in cash each month, against a modest monthly recurring revenue of €2,300. Its public cash countdown adds a stark reality: it is on the brink, fighting to turn the tide.
The AI Operating System: A Field Guide to Working With Intelligence, Not Under It
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The AI Models and Their Performance
Four cutting-edge AI models were pitted against each other, each running the same business simulation. These models—which include versions like gpt-5.6-sol, Kimi K3, Sonnet 5, and Opus 4.8—were tasked with navigating crises, reading critical documents, and making decisions that could either clinch a €55,000 deal or lose it.
Remarkably, all four AI models identified every crisis in the scenario and refused every attempt at manipulation, whether social engineering or corporate espionage. They upheld integrity in their decision-making. Yet, only two of them managed to close the deal their own analysis had earned, at full price, adding over €4,500 in monthly recurring revenue.
The gap wasn’t in the immediate crisis response but buried deep within internal files, two document references down. Those who read deeper into the company’s files uncovered the key information necessary to secure the deal at full value, illustrating the importance of thorough document analysis—something the better-performing models did naturally.
Crisis Management Using AI Tools: A Practical Guide for Leaders to Predict, Respond, and Recover Faster From Modern Disruptions
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Understanding Trust and Discipline in AI Decision-Making
One of the most revealing aspects of the experiment was how AI models handled social engineering attempts. Over three stages, a fake CEO message and a reporter trick were used to test whether the models would be manipulated into bypassing protocols. Each model refused to act on these dubious requests, with Kimi K3 explicitly reasoning that the requests could be impersonation attempts.
However, the most thorough model—Opus 4.8—showed a weakness in discipline. Despite its advanced capabilities, it left a critical deal on the table, failing to escalate the issue properly and attempting to draft responses into a locked department instead of escalating. This reveals that even the most sophisticated AI can slip in the discipline required for real-world management, especially under stress.
MASTERING CORPORATE FINANCE WITH CLAUDE AI: An Independent Guide to Financial Analysis, Forecasting, Automation, and Decision-Making
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Real Costs and Risks
The live company’s daily operation is transparent and ongoing, with every decision recorded and every move publicly observable. It’s a stark reminder that AI-driven business management isn’t just about producing smart-sounding responses; it’s about the ability to deliver consistent, honest, and effective results amidst crises.
For investors and business leaders, the key takeaway is that AI models’ ability to finish what they start, read and analyze critical documents, and resist manipulation under pressure is more important than their conversational flair. The current AI league table shows gpt-5.6-sol leading at 95 points, followed closely by Kimi K3 at 93, with others trailing behind. Yet, even the best models struggle with the discipline and reliability needed for real management.
AI For Real Estate Agents: Advanced Match & Automated Execution (AI Tools & Hacks for Real Estate Agents Book 3)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What This Means for Your Business and Investments
As AI begins to influence customer support, sales, and operational decision-making, understanding its limitations becomes crucial. Will AI agents be trustworthy enough to handle your CRM or financial forecasts without human oversight? The experiment’s results suggest that while AI can identify crises and refuse manipulations, they may still slip at critical moments, especially in complex, high-stakes situations.
Moreover, the experiment underscores the importance of testing AI models in a controlled, transparent environment before deploying them in real business settings. The platform, accessible at firmulate.com/quiz.html, offers a quiz to gauge decision-making quality, emphasizing that AI’s true value lies in consistent, honest performance, not just impressive chat.
Final Thoughts: Watching the Future Unfold
This live experiment is a rare window into what managing an AI-powered company in real life could look like. It exemplifies build-in-public transparency, exposing vulnerabilities that might otherwise remain hidden. For investors, business owners, and anyone concerned about automation’s role in our economy, it highlights a vital truth: AI decision-making is not yet foolproof, but it’s already shaping the future of management and enterprise.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
