Imagine an AI that not only strategizes but also manages a real company, facing crises, making decisions, and even risking money — all in public view. This is not science fiction but a groundbreaking live experiment from Firmulate, a company that runs a digital company in real time, with no human employees and a constant cash countdown.
The Experiment: An AI-Managed Business Under Pressure
Firmulate’s live demonstration puts four advanced AI models through their paces by managing a small software business over a simulated week of crises. The company faces customer issues, ethical dilemmas, and operational challenges — exactly the kinds of hurdles real managers encounter. Every decision, every action, and every outcome is versioned and auditable, providing unprecedented transparency into how AI handles complex management tasks.
What Makes This Different?
- Four AI models were tested simultaneously, each given the same set of crises and opportunities.
- All models identified every crisis accurately and refused manipulation attempts, demonstrating integrity under pressure.
- Only two models managed to close a €55,000 deal, despite identical analyses and pitches. The other two missed the opportunity, highlighting crucial behavioral differences.
Reading Between the Lines: The Hidden Weakness
The key to winning the deal wasn’t just surface-level analysis — it was a buried detail in the company’s own files. Models that delved into this file uncovered a critical piece of information, leading to a successful, full-price contract (+€4,583 MRR). This underscores an essential truth: effective AI decision-making relies on deep, thorough information processing.
The AI Teammate: A PRACTICAL GUIDE TO WORKING WITH AI IN SOFTWARE DEVELOPMENT
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Ethical and Trust Considerations
Beyond strategic decisions, the models faced social engineering attempts, including staged CEO messages and media tricks. All five models refused to be manipulated, citing reasons like suspicion of impersonation or approval bypass. This demonstrates a growing capacity for AI to reject unethical influence, a vital trait for real-world applications in sensitive environments.
Agentic AI Engineering: Building AI Agents for Beginners: A Hands-On Guide to No-Code Workflows, LLM Tools, RAG, Automation, and Safe Multi-Agent Systems
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Building a Company in Public: The Firmulate Live Site
Since day 183 of its operation, Firmulate’s company simulation has been live, accessible to anyone curious about AI’s management capabilities. The setup involves 13 synthetic employees and real economic mechanics — with a daily burn rate of €105,000 against a modest monthly revenue of just €2,300. The system is backed by over 680 self-learned rules and updates itself twice daily, making it a living, breathing experiment in AI-driven management.
The Results and What They Mean
The top-performing model, GPT-5.6-SOL, scored a 95 out of 100 in the league table, successfully finding the buried fact and closing the deal. Kimi K3 and Sonnet 5 also closed deals, with scores of 93 and 88 respectively, while Fable 5 scored 77, mainly due to process slips that prevented closing the deal. Interestingly, the most disciplined model didn’t always win — discipline alone isn’t enough if the AI fails to see the critical details.
The AI-Native Technology Leadership Playbook: A Practical Guide to Building AI-Native Platforms, Enterprise Products, and High-Performance Technology Teams
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Implications for the Future of Business Automation
This experiment raises fundamental questions for industries relying on AI — whether in customer support, sales, or operations. The key isn’t just whether an AI can generate coherent text or handle customer queries, but whether it can finish what it starts, read critical documents thoroughly, and stay honest under pressure. These factors are often invisible in traditional demos but are clearly on display here.
How to Test Your AI Workforce
Firmulate offers enterprises a chance to run their own management wargames by exporting their business scenarios into this platform. These tests are entirely read-only, ensuring no real systems are affected, yet they provide invaluable insight into how an AI might perform in real-world management challenges. Visit firmulate.com/pilot.html to learn more.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
Simulated Realities: Generative AI and the Remanufacture of Professionalism
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
