What if your business could be run by artificial intelligence — and you could watch it happen in real time?
Imagine a company with no employees, yet constantly navigating crises, making decisions, and even negotiating deals — all while its future hangs in the balance. This is not science fiction but a groundbreaking live experiment that puts AI through its paces as a management team in a miniature, fully transparent company.
As an affiliate, we earn on qualifying purchases.
The unprecedented experiment: running a business in public
At the core of this adventure is Firmulate, a platform that emulates a small software business with real money mechanics, but with a twist: it’s run entirely by AI models that act as the company’s decision-makers. This setup allows for a transparent, real-world test of AI’s capabilities in complex, crisis-laden environments.
Four advanced AI models, including the notable GPT-5.6-sol and Kimi K3, are tasked with running the same company through its worst week. Every decision — from handling customer crises to negotiating contracts — is meticulously recorded and versioned, enabling detailed analysis and comparison.
Key findings from the AI management trial
- All four models identified every crisis and refused attempts at manipulation — whether fake CEO messages or behind-the-scenes tricks.
- While they recognized issues, only two models managed to close a critical €55,000 deal, earning an extra €4,583 in monthly recurring revenue (MRR).
- The decisive advantage came from the models that read and utilize information buried two document references deep within the company’s files, revealing a hidden opportunity that others missed.
- Interestingly, the models that were more disciplined or thorough did not always succeed; for example, Opus 4.8, despite analyzing deeply, left a deal unexecuted due to discipline slips, illustrating that thoroughness alone isn’t enough to guarantee success.
The experiment underscores a vital point for any business: AI’s ability to spot critical hidden opportunities and act decisively can be the difference between closing a deal or losing it.
Surviving crises and ethical challenges in a digital management environment
Beyond decision-making, the experiment tested how AI handles social engineering — fake CEO messages and media tricks designed to manipulate human managers. Remarkably, all five models refused to participate or be duped, citing concerns like impersonation risks or approval bypasses, demonstrating a cautious, security-minded approach.
This kind of resistance to manipulation is crucial, especially as AI begins to integrate more deeply into business functions touching real people and sensitive data.
The live company: a microcosm of AI in the real world
Unlike typical AI demos, this experiment runs within a live, publicly accessible environment at firmulate.com/live.html. It features 13 synthetic employees, each governed by a set of over 680 self-learned rules, operating against a backdrop of real money mechanics with a monthly burn rate of €105,000 against a modest €2,300 MRR.
Every workday, the system updates and version-controls its decisions, providing a real-time window into how AI manages financial and strategic pressures. This transparency offers a unique opportunity for businesses and researchers to observe AI’s decision-making in a high-stakes setting.
Lessons from the trenches
The most thorough model, Opus 4.8, with over 80 learned rules and deep analysis, was the last to close a deal — and left opportunities on the table due to discipline lapses, such as relapsing into internal write attempts instead of escalation. This highlights that even advanced models need balanced discipline and process adherence to succeed consistently.
For business leaders exploring AI integration, the question is no longer just about chat quality or conversational finesse. It’s whether AI can execute tasks, stay honest under pressure, and deliver measurable value — aspects critical to real-world applications like CRM, support, and forecasting.
A living scoreboard and open challenges
The live experiment not only shows AI’s potential but also invites engagement. Visitors can participate in a quiz with 242 real management decisions, or run their own business scenarios against a read-only export of the system, ensuring the experiment is accessible and educational.
As AI management models advance, this transparent, build-in-public approach offers an invaluable glimpse into what it takes for AI to truly operate in the complex, unpredictable world of business.
Key takeaway: AI can recognize crises, resist manipulation, and even close deals — but success depends on discipline, thoroughness, and understanding buried information. Watching this experiment unfold reveals how close we are to AI managing real companies, with all the risks and opportunities that entails.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
