Imagine a sports team that’s playing every game live for fans to watch, making their moves in real time, yet losing more than they earn with each match. This isn’t a fantasy—it’s the reality of a pioneering AI-driven software company that’s openly streaming its daily struggles and decisions, all in the name of innovation. For sports enthusiasts familiar with watching underdog teams fight against the odds, this story of a company fighting for survival in full public view offers a compelling parallel.
Building in Public—With a Twist
The company, operating with no human employees, relies on 13 synthetic ’employees’ powered by advanced AI models. Each day, it runs through simulated crises, customer interactions, and decision challenges. The goal? To see whether AI can manage complex business scenarios under real-world pressures, all while being transparent about every move.
This live experiment is available for anyone to observe at firmulate.com/live.html. It’s a transparency experiment like no other: every decision, every crisis, and every rule learned by the AI models is documented and versioned. The stakes are high, with the company burning €105,000 every month against a modest €2,300 in monthly recurring revenue (MRR). This public cash countdown underscores the ongoing challenge: can AI manage a business and make it profitable?
AI business decision simulation software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The AI Models: Competing in a High-Stakes League
The experiment pits four frontier AI models against each other, each tasked with running the same small software firm through its worst week. Every model faces the same crises, customer demands, and temptations—like attempts at manipulation or social engineering. The top scorer, gpt-5.6-sol, achieved a score of 95 out of 100, successfully uncovering a hidden opportunity that led to a €4,583 MRR increase. Kimi K3 closely followed with a score of 93 and the best discipline, also closing a deal. Meanwhile, Sonnet 5 scored 88, and Fable 5 lagged at 77, despite having the best rules discipline.
What’s remarkable is that all four models recognized every crisis and refused every attempt at manipulation, including fake CEO messages and reporter tricks. Only two signed a deal worth €55,000 based on their own analysis—showing that AI can identify opportunities, but not all succeed in executing them fully.
The Hidden Weakness and the Power of Reading Files
The decisive advantage, however, wasn’t in surface-level decision-making but hidden deep within the company’s internal documents. Models that delved into these files uncovered critical information that led to closing the deal at full price—a difference worth over €4,500 in monthly revenue. This emphasizes the importance of thorough information reading and analysis, even in automated decision-making.
Social Engineering and AI Integrity
In scenarios testing social engineering—a common tactic in business and security—the AI models consistently refused to be duped. Fake CEO messages, background queries, and staged reporter questions all failed to sway their decisions. Kimi K3 justified its resistance by treating these requests as potential impersonation, showcasing a cautious and disciplined approach.
The Real-World Stakes and Daily Struggles
The live company’s setup is extraordinary: no employees, just AI models that learn rules and adapt each day. Every workday, the entire operation is versioned, and performance is publicly accessible. The company burns through €105,000 monthly, trying to turn a minimal €2,300 in revenue into a sustainable business.
This isn’t just a tech demo—it’s a real, ongoing story of an AI trying to run a business under the same conditions as a struggling startup. The experiment provides a stark view of what AI can—and cannot—do in high-pressure situations, revealing weaknesses in discipline or insight that even the most advanced models exhibit.
Implications for the Future of AI in Business
Why does this matter beyond the tech world? If AI agents are going to interact with your customer relationship management systems, support queues, or financial forecasts, the question isn’t whether they can produce eloquent chat responses. Instead, it’s whether they can follow through, verify critical information, stay honest under pressure, and ultimately deliver useful work that truly benefits the business.
With every decision documented and auditable, this experiment offers a transparent test bed for AI’s readiness to handle real-world business challenges—something traditional demos can’t replicate.
This open experiment shows that AI can identify crises, resist manipulation, and even close deals—if it reads the right information. But discipline, thoroughness, and honesty remain challenges as AI enters the business arena. For companies considering AI integration, watching this live story unfold offers invaluable lessons on potential strengths and pitfalls.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
