Imagine watching an AI-led company fight for survival every single day — without employees, yet with real money at stake. This is not science fiction; it’s the live experiment by Firmulate, a company that turns AI models into synthetic executives navigating crises, opportunities, and threats in real time. For gamers and tech enthusiasts alike, this is a behind-the-scenes look at how artificial intelligence is tested in the brutal arena of business, revealing not just what AI can do, but what it can’t — yet.
The Experiment: Turning AI into a Company Executive
Firmulate’s latest live experiment puts four frontier AI models — including versions of GPT-5.6 and Kimi K3 — through the ultimate test: managing a small software company during its worst week. Every decision, crisis, and temptation is meticulously documented and versioned, creating a transparent and auditable record of how these AI ‘executives’ perform under pressure. The company itself is real — with 13 synthetic employees, a €105,000 monthly burn rate against only €2,300 in monthly recurring revenue, and a publicized cash countdown at firmulate.com/live.
As an affiliate, we earn on qualifying purchases.
Results That Speak Louder Than Demos
All four AI models successfully identified and responded to every crisis, refusing manipulation attempts — including an elaborate social engineering test involving fake CEO messages and a reporter trick. Remarkably, only two of these models managed to close the €55,000 deal their own analysis identified as the right choice. The others, despite diagnosing correctly and making the same pitch, left the deal unclosed. It’s a stark reminder that AI’s decision-making isn’t just about understanding — it’s about execution and discipline.
The Hidden Weakness: Reading the Files
Digging deeper, the experiment uncovered a crucial flaw: the decisive advantage in closing the deal was tied to reading specific internal documents — references buried two layers deep in the company’s files. Models that read these files won the full-price deal, worth an additional €4,583 in monthly recurring revenue, whereas those that missed this buried fact failed to secure the deal at all. This highlights a vital aspect of AI capability: access to complete and accurate context is often the key to success in complex decision environments.
Resisting Social Engineering and Ethical Tests
The experiment also tested how these AI models handle social engineering — attempts to manipulate decisions through fake CEO messages and background requests. All five models refused to act on these, with Kimi K3 explicitly treating the requests as potential impersonation or approval bypasses. This demonstrates a growing awareness in AI systems around security and ethical boundaries, even when facing escalating pressure.
The Live Company: A Daily Battle for Survival
Firmulate’s live company is more than an experiment; it’s a real-time window into the challenges of managing AI in a business context. Every weekday, the company updates its decision records, with over 680 learned rules guiding its synthetic employees. Despite this, it is losing money — burning €105,000 each month against a tiny revenue stream, with its public cash countdown adding a sense of urgency and transparency.
Lessons for the Gaming and Tech World
For gamers and interactive entertainment fans, this story offers a fascinating parallel: AI as a strategic player that must read the board, anticipate crises, and act ethically under pressure. It’s not just about generating convincing dialogue or impressive graphics; it’s about making decisions that hold up under scrutiny and delivering consistent results. The experiment underscores that in AI-driven environments, the ability to finish what you start, to read hidden information, and to resist manipulation, is what separates the truly effective from the merely impressive.
The Path Ahead for AI-Managed Companies
While current models are capable of identifying crises and refusing manipulative tactics, the gap remains in the discipline and follow-through needed for real-world business success. The experiment shows that even the most advanced AI can slip, especially under pressure or when relying on incomplete information. This is a critical insight for companies considering AI for operational roles: understanding AI’s strengths and weaknesses is essential before integrating it into decision-critical processes.
Get Involved: Watch and Learn
If you’re curious to see this experiment in action, visit firmulate.com/live. Watch the live updates, explore the decisions made by each model, and get a firsthand look at how AI manages business crises in real time. For a more interactive experience, test your management intuition at firmulate.com/quiz.html — and see if you can tell which AI model is making which decision.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
