Watch an AI Run a Company in Real Time — and See Its Flaws in Action

Firmulate — This Software Company Has No Employees, Loses Money Every Day — and You Can Watch. — Live on firmulate.com.

Imagine if your favorite travel guide or outdoor gear brand was run not by humans, but by artificial intelligence — making decisions, facing crises, and sometimes even failing at the worst moments. Now, what if you could watch that happen live? Welcome to the world of Firmulate, where a small, virtual software company is battling for survival on a public stage, with every move and mistake on display.

The Live Experiment: AI as a Company CEO

At the heart of this experiment is an unusual company: a simulation run by four advanced AI models, each tasked with managing a tiny software business. This isn’t just a demo; it’s an ongoing, transparent test of how AI handles real-world business challenges. Watch at firmulate.com/live.html.

How It Works

The AI models are given the same difficult week: same customers, same crises, and the temptation to cut corners. Each model’s decisions are recorded and publicly available, giving an unprecedented window into AI decision-making in a high-stakes environment. Every day, the system is versioned, with over 680 self-learned rules guiding the process.

The Results So Far

All four models identified and responded to every crisis, demonstrating a high level of situational awareness.
Only two models managed to close a €55,000 deal based on their analysis, despite identical conditions and pitches.
The key difference was found in their ability to read deeper into internal documents — the models that examined their own files uncovered a crucial piece of information that sealed the deal, adding €4,583 MRR.
Social engineering attempts, like fake CEO requests or reporters’ tricks, were universally refused, thanks to built-in safeguards.

Why Does This Matter?

This experiment is more than just a tech showcase. For businesses relying on AI to manage customer relationships, forecasts, or support, the questions are clear: Will it stay honest under pressure? Will it find hidden opportunities? And most importantly, will it finish what it starts?

In one instance, an AI with thorough analysis (Opus 4.8, the most disciplined participant) failed to close a deal because it left the opportunity unexecuted, slipping into a disciplinary lapse. Meanwhile, the winner, GPT-5.6, not only found the concealed fact but also closed the deal, earning full recognition for its performance.

The Company Behind the Curtain

This isn’t a fictional story. The company running this live experiment has no employees—just 13 synthetic “workers” and a complex web of rules and mechanics designed to mimic real business operations. It’s losing €105,000 every month against a modest €2,300 monthly recurring revenue — a stark illustration of how AI-driven companies might operate in their earliest, most fragile stages. The public cash countdown underscores the urgency: survival hangs by a thread, and every decision counts.

The Broader Implications

For outdoor enthusiasts, travelers, or anyone interested in technology’s frontier, this experiment reveals something vital: AI decision-making in business is not just about writing well or sounding convincing. It’s about integrity, attention to detail, and the ability to finish what’s started, especially under pressure. As AI begins to touch your everyday life — from customer service to personal assistants — understanding its strengths and vulnerabilities becomes crucial.

Try It Yourself

Businesses and curious minds can run their own version of this AI “war game” against their real operations, without risking their actual data. It’s a safe, transparent way to see if an AI can handle the kind of crises your company faces daily, and whether it can be trusted to follow through. Details and access are available at firmulate.com/pilot.html.