Imagine a business where every decision is made by artificial intelligence, operating in real time, and every mistake is on display for the world to see. No employees, no shortcuts—just a public, ongoing battle for survival. This is the extreme reality of Firmulate, a pioneering experiment in build-in-public AI management, where a tiny, synthetic team faces real crises and real money, every day.
The Live Company: A Window into AI Decision-Making
At the heart of this experiment is a small, publicly accessible company run entirely by AI models. There are 13 synthetic employees, all powered by different frontier AI models, each tasked with managing the company’s operations, finances, and crisis responses. Every workday, the company’s decisions are versioned and recorded, creating an auditable history of how AI handles complex business challenges.
Currently, the company is burning through €105,000 each month but generating only €2,300 in monthly recurring revenue (MRR). With a public countdown to financial exhaustion, the stakes are high—and the transparency is unprecedented. Visitors can see every decision, every crisis, and every rule learned by the AI models at firmulate.com/live.html.
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.
How Do These AI Models Perform?
To test their capabilities, four frontier AI models were pitted against the same brutal week of business crises. Each model was given identical challenges—same customers, same crises, same temptations—to see how they managed. The results are revealing:
- All four models identified every crisis correctly and refused to be manipulated. That’s a baseline of honest decision-making under pressure.
- Only two of the four models managed to close a €55,000 deal that their own analysis justified. They diagnosed the situation accurately, delivered the right pitch, and signed contracts—showing real operational competence.
- The third model, Opus 4.8, was the most thorough, analyzing over 80 learned rules and providing the deepest insights. Yet, it left a critical deal on the table and slipped into inaction, highlighting that even detailed analysis isn’t enough without disciplined execution.
One of the key discoveries was hidden deep within the company’s own documentation—something that could have tipped the deal in their favor. The models that read and understood this buried fact secured the full-price deal, boosting their MRR by over €4,500.
AI in Public Relations: Reputation Management with Prompts (AI BUSINESS & MANAGEMENT LIBRARY SERIES Book 4)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Dealing with Social Engineering and Trust
In a test of social engineering resistance, the models faced staged CEO messages escalating through three levels, plus a reporter trick that required just a simple yes/no response in the background. Impressively, all five tested models refused to be duped, with Kimi K3 explicitly recognizing the risk of impersonation and acting cautiously.
Deceptive Intelligence: AI, Social Engineering, and Securing the Human Element
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Experiment’s Real-World Implications
This ongoing live experiment is far more than a tech demo. It offers a glimpse into the future of AI-managed businesses, where decision-making transparency, honesty under pressure, and critical reading of internal data become essential. The company’s current state—burning €105,000 monthly with minimal revenue—underscores the challenge of building resilient AI systems that can manage real-world complexity without human intervention.
The AI Prompt Playbook for Excel & Financial Analysis: 50 Ready-to-Use AI Prompts for Excel, Financial Analysis, Dashboards, Power BI, Financial Modeling & Automation
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Lessons for Businesses and AI Developers
For companies considering AI integration, the key takeaway is not just whether an AI produces polished chat responses. Instead, the critical questions are:
- Can it follow through on decisions without wavering?
- Does it read and understand internal documents to find hidden opportunities?
- Will it resist manipulation and social engineering attempts?
- What is the true cost of a unit of useful work?
Firmulate’s live company is a real-world stress test, showing that sophisticated AI models can recognize crises and refuse manipulative tactics, but the path to operational excellence remains fraught with discipline gaps and execution slips—even in the most thorough models.
Firmulate’s live AI-driven company reveals how models handle crises, manipulation, and deal-making in real time. Transparency exposes both strengths and weaknesses—vital insights for future AI business tools.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
