Watch an AI-Run Business Fight for Survival in Real Time

This Is Not a Simulation — It’s a Live Battle for Business Survival

Imagine a company that has no human employees, yet faces the same crises, temptations, and tough decisions as any real business. Now imagine you can watch this company every day, live, as it struggles to stay afloat. That’s the reality with Firmulate’s experimental AI-driven company, a rare window into how artificial intelligence manages real-world business challenges in real time.

The Live Experiment: An AI-Driven Company Under Pressure

At the heart of this experiment is a small software company run entirely by AI models, each acting as a team of synthetic employees. These models are tasked with managing day-to-day operations, handling crises, and pursuing deals — all while losing €105,000 each month against a revenue of just €2,300.

What makes this setup extraordinary is its transparency and accountability. Every decision made by the AI models is versioned and auditable, providing a clear view of their reasoning and actions. The entire operation is accessible at firmulate.com/live, where you can watch the company in action, see its cash countdown, and observe the decision-making process unfold in real time.

The Models and Their Performance

• GPT-5.6-sol scored the highest with a 95/100, successfully uncovering hidden opportunities in company files that led to closing a €55,000 deal, boosting monthly recurring revenue (MRR) by €4,583.
• Kimi K3, a newcomer, scored just slightly below at 93/100, also closing the same deal but with the clearest discipline among all models.
• Sonnet 5 and Fable 5 followed, with scores of 88 and 77 respectively, each closing deals but slipping on process discipline and decision accuracy.

Despite their successes, the models face fundamental challenges. For instance, Opus 4.8, with the most comprehensive analysis (over 80 learned rules), left a crucial deal on the table due to a lapse in escalation discipline. The models also refused social engineering tricks, such as fake CEO messages or reporter manipulations, demonstrating a basic level of ethical robustness.

Why Does This Matter for Business and AI?

In a world where AI assistants are increasingly integrated into customer service, sales, and operations, the question isn’t just whether they can generate convincing chat responses. Instead, the critical issues are:

• Can the AI follow through on complex tasks without slipping into shortcuts or unethical behavior?
• Will it read and understand critical internal documents that could unlock better decisions?
• Will it stay honest under pressure, resisting manipulation or social engineering attacks?

The real-world performance of these models in a CEO’s shoes offers a stark contrast to what we see in scripted demos or chatbots. The experiment shows that an AI can recognize crises, refuse manipulative requests, and even uncover hidden opportunities, all while managing a business that is bleeding cash daily.

The Cost of Failure and the Path Forward

Right now, this real company is losing €105,000 every month against a tiny revenue stream. Its public cash countdown makes visible the urgency of the situation. Every decision, every rule learned, every process slip and success is part of an ongoing story that anyone can follow, learn from, or even challenge.

For business leaders and technologists alike, the key takeaway is clear: AI’s ability to reliably complete complex, ethical, and strategic work in a high-pressure environment is still a work in progress. But with experiments like this, we can observe, learn, and adapt — taking the guesswork out of AI integration and paving the way for systems that truly augment human decision-making, not just simulate it.

See It in Action

Visit the live site to watch the company’s daily battle unfold. Explore the decisions, read the real-time quotes, and get a firsthand look at how AI is managing a business in its most raw and revealing form.

Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html

Powered by Thorsten Meyer AI