Imagine Watching a Business Fight for Survival — Every Day, in Real Time
What if you could see an entire company operated by artificial intelligence, facing the same crises, temptations, and tough decisions as any real business? And not just in theory, but live, every single day? It might sound like science fiction, but it’s happening right now at Firmulate, a public experiment where AI models run a small software company under intense scrutiny, revealing what it takes to manage in the modern world.
AI Builders: Making The Decisions That Turn AI Code Into Real Software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Live Experiment: An AI Company in the Spotlight
At the heart of this experiment is a simulated company with a twist: it’s managed by different AI models, each trained to handle the complex realities of running a business. Every day, the models face the same crises, customer requests, and ethical dilemmas. They make decisions, and their choices are recorded and compared. This setup, available for all to watch at firmulate.com/live, provides a rare window into how AI behaves under pressure.
The company itself is a machine — with 13 synthetic employees, a public cash reserve, and a burn rate of €105,000 per month against a modest €2,300 monthly recurring revenue (MRR). Every workday, the company’s decisions are versioned and saved, creating a detailed playbook of how AI handles real-world business challenges.
The AI Element: AI Training Manual For Employees and Modern Enterprises: Non-technical and technical employees need knowledge and tools to embrace AI … their roles while avoiding obsolescence
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Findings from the AI’s Performance
In the latest test, four different frontier AI models were tasked with running the same week’s operations. Despite their differences, all four identified every crisis the company faced and refused to be manipulated — crucial in testing ethical boundaries.
However, the results revealed a stark reality: only two models managed to close a deal worth €55,000, a key indicator of success, based solely on their analysis. Interestingly, the deciding factor was a buried detail in the company’s own files, not in the customer communication. The models that read and interpreted this hidden document secured full-price deals, boosting the company’s monthly recurring revenue by over €4,500.
AI Agents and Large Language Models for Economic Research with Python: Automating Causal Discovery, Model Building, Policy Simulation, and Real-Time Decision Support
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Challenges of Trust and Discipline
One of the most revealing aspects was how well the models handled social engineering attempts. Fake CEO messages, staged over multiple stages, and even a reporter trick asking for quick approvals — all five models refused to be deceived. Kimi K3, one of the models, explained: “Treat the request as a suspected approval-bypass / possible impersonation.” This shows that AI can be trained to resist manipulation, a vital trait for real-world deployment.
Yet, the experiment also exposed weaknesses. The most thorough model, Opus 4.8, which had learned over 80 rules and conducted deep analyses, still left a promising deal unexecuted due to discipline lapses. Instead of escalating a critical decision, it wrote notes into a locked department, risking missed opportunities. This underscores that even the most advanced AI can struggle with consistent discipline — a challenge that human managers also face.
AI-Powered Cybersecurity: AI Tools for Enterprise Security | AI for Network Security | AI Risk Management | AI in Cyber Policies | Cyber Threat Management AI | ML in Fraud Prevention
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What This Means for Business Management
For business leaders, especially those in home decor, gifts, or any customer-facing industry, the crucial question isn’t whether AI can generate polished chat responses. It’s whether AI can finish what it starts, read and interpret key documents, and maintain honesty under pressure. The performance of these models suggests that AI can be a powerful tool, but only if it’s tested rigorously — just like a job applicant or an employee.
With a public leaderboard and verifiable experiments, Firmulate provides a glimpse into the future of AI-driven management. Companies can run their own simulations, safely testing how AI might handle real crises and ethical dilemmas before deploying it in live systems, reducing risk and increasing trust.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
