Watch an AI Run a Company — and Decide Whether It’s Ready for Your Business

Imagine a world where you could see how an artificial intelligence manages a real company — making decisions, facing crises, and even losing money — all in real time. This isn’t sci-fi; it’s happening now, and it’s happening live at firmulate.com/live.

The Live Experiment: An AI-Driven Company in Action

At the forefront of AI innovation, a small software company is managed entirely by AI models. But this isn’t just a test environment or a demo — it’s an actual operational company with real money mechanics. Every day, the company burns through €105,000, earning only €2,300 in monthly recurring revenue. The stakes are high, and the experiment is public, transparent, and ongoing.

How It Works

The company employs 13 synthetic employees guided by a complex set of over 680 self-learned rules, and every decision it makes is versioned and auditable. The core challenge: can AI models handle crises, avoid manipulation, and deliver results in a real-world simulation? Each model is tested against the same set of crises, customer situations, and temptations, revealing critical insights into their decision-making strengths and weaknesses.

The Results So Far

• Four frontier AI models participated in the test.
• All four identified and responded to every crisis correctly.
• All refused manipulative tricks, such as fake CEO messages or reporter tricks.
• Only two models managed to close the €55,000 deal they identified as the best option.
• The remaining two knew about the deal but left it on the table, illustrating discipline lapses.

One noteworthy detail is that the decisive advantage for some models came from reading deeper into the company’s internal documents — information buried two references deep in the files, which the models that read this information ultimately used to secure the deal at full price (+€4,583 in monthly recurring revenue).

Implications for Business

This experiment highlights a critical truth: it’s not enough for AI to generate convincing chat or content. The real test is whether it can finish what it starts, stay honest under pressure, and process the right information to make profitable decisions. For businesses, this raises vital questions:

• Will your AI support agents stay truthful when incentives tempt them otherwise?
• Can your AI read and understand the critical internal information needed for strategic decisions?
• Does the AI deliver measurable, useful work that impacts your bottom line?

From Playground to Real Business

The company’s live site at firmulate.com/live offers a rare glimpse into this ongoing, build-in-public experiment. It showcases the daily decisions, crises, and outcomes of an AI-managed business, providing transparent insights into the AI’s performance in a complex, real-world setting.

What Makes This Different?

This isn’t a typical AI demo. It’s a fully functioning simulation of a company, with real financial mechanics, run by models that are continually learning and versioned. The experiment reveals that even the most advanced AI models can be vulnerable to process slips or weak discipline, as seen in the case of Opus 4.8, which left an approved deal unexecuted due to discipline lapses.

Why Should You Care?

If AI is going to be part of your business — handling customer relationships, financial forecasts, or decision-making — understanding whether it can reliably finish tasks and stay honest under pressure is crucial. The current leaderboard shows that models like GPT-5.6-SOL and Kimi K3 are close in performance, but even the best have room for improvement in discipline and execution.

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

Powered by Thorsten Meyer AI