The Future of AI in Business? Watch a Company Struggle Live
Imagine an AI-powered company that operates without employees, fights for every penny, and is open for all to see. This isn’t a sci-fi story — it’s real, ongoing, and happening right now. Welcome to the world of Firmulate, where a small, virtual company is living its worst week, and you can watch every decision unfold.
AI decision-making software for business
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Inside the Live Experiment
Firmulate runs a groundbreaking experiment: it models a tiny software business using artificial intelligence, with 13 synthetic employees making real decisions about real money. Every workday, the system is versioned, audited, and transparent, giving a rare glimpse into how AI performs under pressure.
The goal? To test whether AI can handle crises, resist manipulation, and make sound decisions in a realistic business environment. The experiment features four leading AI models, each challenged with the same set of scenarios, crises, and temptations.
Decisive Results in a High-Stakes Test
Despite all four models accurately identifying every crisis and refusing every manipulation attempt — including sophisticated social engineering — only two managed to close a €55,000 deal based on their own analysis. The other two, even with the same diagnosis and pitch, left the deal unexecuted, illustrating the importance of discipline and follow-through in AI decision-making.
One of the top performers, GPT-5.6, not only closed the deal but also uncovered the company’s hidden vulnerability buried deep in internal files, which, if exploited, could have secured an additional monthly recurring revenue of over €4,500.
The Human and AI Intersection
Throughout the week, the system faced numerous crises: customer issues, internal conflicts, and social engineering attacks like fake CEO messages and reporter tricks. All five AI models refused to be manipulated or bypass controls — a sign of robust integrity.
Interestingly, the model Kimi K3, which runs without an effort parameter, demonstrated the cleanest discipline, closing the deal without attempt to game the system, while others faltered or left opportunities on the table.
The Stakes and the Bigger Picture
This experiment isn’t just about AI performance metrics. It exposes how AI agents could impact your business operations — whether in customer support, CRM, or forecasting. The question isn’t just “does it write well?” but rather, “does it finish what it starts? Does it read your files? Does it stay honest under pressure? And what is the real cost of useful work?”
Currently, the live experiment at firmulate.com/live offers a front-row seat to this ongoing story. Watching the company operate in real-time, with decisions versioned and auditable, provides a new level of transparency into AI’s capabilities and limitations in complex scenarios.
The Bottom Line: Building Trust in AI Workforce
As AI increasingly touches core business functions, understanding its behavior in high-pressure situations is critical. The Firmulate experiment shows that AI can be robust against manipulation, but discipline — like closing deals rather than leaving them unrealized — remains a challenge.
This live, build-in-public approach highlights an essential truth: testing AI in realistic, crisis-filled environments is vital before trusting it with real business processes. For startups and established companies alike, the question becomes whether AI can truly be a reliable partner, or if it still needs human oversight.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
