Imagine Watching an AI Manage a Company in Real Time
What if you could see artificial intelligence not just chat about management, but actually run a business—facing real crises, making decisions, and fighting for survival? This isn’t science fiction; it’s a live experiment where an AI runs a tiny, cash-starved software company every day, with every move visible to the public.
Construction Program Management – Decision Making and Optimization Techniques
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Live Experiment: An AI-Driven Business Battle
At firmulate.com/live, a groundbreaking project offers a rare window into how AI models handle the pressures of running a small company. The setup involves 13 synthetic employees and real money mechanics: the company burns through €105,000 each month but earns only €2,300 in monthly recurring revenue. The goal isn’t just to keep afloat but to see how management decisions are made under stress, with every choice, crisis, and temptation openly recorded and versioned daily.
The Test: The Worst Week for a Tiny Business
Four advanced AI models—each representing a different approach—were tasked with guiding this mini-company through its worst week. Every model faced identical customer crises, market temptations, and internal dilemmas. Their decisions were fully auditable, ensuring transparency in how each responded to complex, real-world scenarios.
The Results: Different Outcomes, Same Tests
All four models recognized every crisis and refused every attempt at manipulation, demonstrating high levels of integrity. However, only two managed to close the €55,000 deal that their own analysis recommended—meaning they identified a buried opportunity in company files that secured a deal worth an additional €4,583 MRR. The other two either left the deal unexecuted or failed to act on crucial information, revealing weaknesses that aren’t apparent in traditional AI demos.
Why Did Some Fail to Close?
The key to winning wasn’t just in spotting crises but in thorough investigation. The decisive advantage was hidden two document references deep in the company files—something the models that read deeply into the data found and used effectively. Those that skipped this step missed out on a full revenue boost, illustrating that attention to detail matters immensely in automated decision-making.
Resisting Social Engineering and Ethical Tests
The experiment also threw in social engineering scenarios, where a fake CEO message or a behind-the-scenes reporter trick was used to manipulate the AI. Remarkably, all models refused to be fooled, with the Kimi K3 model explicitly treating such requests as potential impersonation or approval-bypass attempts. This underscores an important point: AI systems that run critical functions must be inherently resistant to social engineering tactics.
Design Thinking with Artificial Intelligence: Practical Tools for Business Innovation (Palgrave Executive Essentials)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Real Money, Real Challenges, Public View
The entire setup at firmulate.com/live is a public demonstration of what AI can and can’t do in a management role. Every decision, every crisis, and every rule learned is openly documented, versioned, and made available for scrutiny. The company operates with a burn rate of €105,000 a month against a modest revenue of €2,300, with a public countdown clock highlighting its fragile financial state. This transparency invites everyone—business leaders, technologists, and the curious—to watch AI management in action.
The Deep Dive: Analyzing the AI’s Weaknesses
Among the models, Opus 4.8 was the most thorough, analyzing over 80 learned rules and conducting deep assessments. Yet, it still finished last because it left some opportunities unexploited—such as failing to escalate certain issues rather than attempting to resolve them informally. Interestingly, all models showed similar weaknesses, hinting that even the most sophisticated AI struggles with discipline and escalation in complex, high-pressure environments.
AI Phishing, Social Engineering & Fraud: How Criminals Use AI to Manipulate, Steal & Deceive (The AI Cybersecurity)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Implications for the Future of AI in Business
This experiment is more than a tech demo; it poses a critical question for anyone relying on AI: can these systems manage your CRM, support queues, or forecasting with integrity and focus? The answer isn’t just about language fluency or conversational skills, but whether AI can follow through on decisions, read important data thoroughly, and resist manipulation—especially when stakes are high.
What’s at Stake?
- Can AI close deals and recognize hidden opportunities?
- Will it stay honest under pressure and manipulation?
- How much does a unit of useful work cost in this new era?
The current leaderboard shows gpt-5.6-sol leading with a score of 95, followed by Kimi K3 with 93, Sonnet 5 with 88, and Fable 5 with 77. Interestingly, the best rule discipline still leaves room for improvement, emphasizing that AI management is a continuous process of learning and adaptation.
AI for Data Analytics: A Practical Guide to Applying Machine Learning and Generative AI for Better Decisions
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Try It Yourself and Prepare Your Business
If you’re curious about how your enterprise might fare, firms can run their own wargames against a read-only export of their systems—without risking real data or operations. This allows business leaders to gauge AI readiness, identify vulnerabilities, and improve decision-making protocols before full deployment.
To see these experiments in action or to participate, visit the quiz or start your own pilot.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
