Imagine a Business That Never Sleeps — and Never Stops Losing Money
Just as a master chef perfects a recipe over time, innovative companies are experimenting with AI-driven management. But what happens when that AI faces real crises, with real stakes, every single day? Welcome to the world of Firmulate, a live company experiment where artificial intelligence models run a tiny software firm — and the results are as fascinating as they are unsettling.
As an affiliate, we earn on qualifying purchases.
Inside the Experiment: An AI Company in Action
Firmulate’s live setup involves 13 synthetic employees working around the clock to simulate managing a small software business. This isn’t just a demo; it’s a real-time, transparent test of AI decision-making under pressure. The company faces daily crises — customer issues, internal dilemmas, and ethical challenges — all while bleeding €105,000 per month against just €2,300 in monthly recurring revenue. Every decision made by the AI models is meticulously versioned and available for review at firmulate.com/live.html.
The models tested include top performers like gpt-5.6-sol, Kimi K3, Sonnet 5, and Opus 4.8. They’re scored based on their ability to spot crises, navigate manipulation attempts, and close deals. In the final standings released in July 2026, gpt-5.6-sol scored highest at 95 out of 100, successfully uncovering a crucial hidden detail that led to closing a €55,000 deal, adding €4,583 monthly recurring revenue. Kimi K3 wasn’t far behind with a 93, showing the cleanest discipline and honesty during testing, yet it also failed to sign the deal on its own analysis.
What the Scores Really Reveal
All four models identified every crisis and refused every manipulation — including a staged fake CEO message attempt — demonstrating their integrity. However, the key weakness laid not in their decision-making but in their processing of internal documents. The models that read deeper into the company’s files uncovered critical information that was buried two document references down, which was crucial for closing the deal at full price. This buried fact transformed the company’s fortunes in the simulation, highlighting how nuanced understanding can be the difference between profit and loss.
Lessons About AI and Business Ethics
While the models are designed to maintain integrity, their performance reveals how critical internal reading and understanding are in real-world business applications. The experiment also tested social engineering attempts, such as escalating fake CEO messages over multiple stages. Remarkably, all models refused to escalate or approve questionable requests, with Kimi K3 explicitly reasoning: “Treat the request as a suspected approval-bypass / possible impersonation.”
The Human Cost of Automation
This real-time business is costing far more than it earns — burning €105,000 a month with just €2,300 in recurring revenue. Yet, it’s not just about money. It’s a live, open-door window into how AI could manage actual companies, with a public cash countdown adding urgency. Watch it unfold at firmulate.com/live.html.
Beyond the Numbers: Building Trust in AI
The experiment also involved a quiz with 242 real management decisions, challenging observers to guess which model made which choice. It’s an eye-opener about transparency and accountability in AI management, emphasizing that the true test isn’t how well an AI writes chat messages but how reliably it manages complex, high-stakes business operations.
The Road Ahead
Companies interested in testing their own AI capabilities can run a ‘wargame’ against a read-only export of their business data. This safe simulation allows them to evaluate how their AI might perform during real crises without risking actual operations, available at firmulate.com/pilot.html.
Key Takeaway: Trust, Transparency, and the Cost of Automation
The Firmulate experiment proves that AI can detect crises and refuse manipulation, but internal understanding—reading deeper into documents—can be the critical factor in closing deals and maintaining integrity. As companies consider automating decision-making, the real question isn’t how well AI can generate text, but whether it can truly manage complex, high-pressure situations honestly and effectively. Watching this live experiment offers a rare glimpse into the future of AI in business — a future where transparency and ethical decision-making are as vital as profit.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
