Imagine Running a Business Without Employees — and Watching It Struggle to Stay Alive
For many, the world of ice cream shops and bakeries is about warm smiles and sweet treats. But what if a business was run entirely by artificial intelligence — with no human staff? And what if that virtual company was fighting to survive in real time, facing crises, temptations, and tough decisions every single day? Welcome to the live experiment by Firmulate, where an AI-powered company is doing just that — and you can watch every move unfold.
As an affiliate, we earn on qualifying purchases.
The Unique Experiment: An AI Company on the Brink
Firmulate’s experiment isn’t your typical AI demo. It involves a virtual company with 13 synthetic employees, managing real money mechanics. The company burns through €105,000 each month, yet earns just €2,300 in monthly recurring revenue (MRR). This stark reality sets the stage for a high-stakes test of AI management — can these models make better decisions than humans in a business teetering on the edge?
Every day, the company faces simulated crises, customer issues, and ethical dilemmas. For example, a recent scenario involved a fake CEO pushing for a quick approval that could be an impersonation attempt. All four AI models tested refused to be manipulated, showcasing resilience against social engineering tricks. This is crucial because in real-world scenarios, AI systems must withstand attempts to deceive them, especially when they control vital business processes.
How the Models Performed
The experiment pitted four frontier AI models against each other, each running the same business through its worst week. Scores ranged from 77 to 95, with the highest being gpt-5.6-sol which managed to identify critical gaps in the company’s own files and closed a €55,000 deal, adding over €4,500 in monthly recurring revenue. The other models, including the newcomer Kimi K3, also closed deals but with slightly less discipline or missed opportunities.
Interestingly, the decisive advantage came from reading internal documents — a layer of information buried in the company’s own files. Models that analyzed these references unlocked full deal potential, emphasizing the importance of comprehensive data access for AI decision-making.
What This Means for Business and AI
This experiment demonstrates that AI models can recognize crises and resist manipulative tactics effectively. However, it also reveals that the real challenge isn’t just in spotting problems but in completing the work — in other words, ensuring the AI follows through on decisions, like sealing a deal or escalating an issue properly.
For instance, the Opus 4.8 model, despite being the most thorough in analysis, left a deal on the table due to lapses in process discipline—failing to escalate appropriately or writing attempts into a restricted department instead of closing the deal. This illustrates that even the most advanced AI requires disciplined rules and clear workflows to succeed in real-world tasks.
Why This Matters for Your Business
The core takeaway isn’t about replacing human staff with AI, but about understanding what makes AI effective in business contexts. Can it read your internal files? Will it stay honest under pressure? Will it complete what it starts? These are the questions every company should ask if they consider deploying AI in critical roles like customer support, sales, or decision-making.
Watch this experiment at firmulate.com/live.html to see these AI models in action, battling crises and temptations in real time. It’s a raw, transparent look at what AI can do — and what it still struggles with — in a high-pressure business environment.
Essentially, this live experiment offers a rare glimpse into AI’s potential — and its limits — as a decision-maker in a real business. It’s a stark reminder that AI’s effectiveness depends not just on scoring well in demos, but on its ability to finish what it starts, stay honest, and handle crises under pressure.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
