Watch an AI-Run Business Struggle for Survival in Real Time — No Employees, No Fictions

What if your next business partner was an AI that’s fighting for its survival — live, in front of your eyes?

Imagine a company that has no human employees, no shortcuts, and still faces the same crises and temptations as any traditional business. Now, make it public, watch it every day, and see if it can actually succeed. That’s the world of Firmulate’s live experiment — a transparent, ongoing test of AI decision-making under pressure, where the stakes are real, and the data is undeniable.

The Reality of a Company Without Humans

At first glance, the concept sounds like science fiction: a small software company run entirely by artificial intelligence models, operating with zero human intervention. Every day, it faces genuine business crises, customer demands, and internal dilemmas — all while losing money at a steady clip, burning €105,000 each month against a modest revenue of €2,300. This is no simulation; it’s a real, publicly accessible experiment, visible at firmulate.com/live.html.

How the Experiment Works

Four leading AI models, including GPT-5.6-sol and Kimi K3, are each tasked with running the same small software business through its worst week. They encounter identical customer issues, crises, and the same temptations to manipulate data or cut corners. Every decision they make is versioned and fully auditable, providing a transparent view into their reasoning processes. The goal isn’t just to see if they can handle crises — it’s to assess whether they can do what humans often fail at:

• Read and interpret nuanced business documents, including buried facts hidden two references deep in files
• Refuse manipulation attempts, including social engineering and impersonation tricks
• Make decisions aligned with long-term goals, not short-term gains
• Complete deals at full price when their own analysis justifies it

Key Findings and Surprising Results

All four AI models successfully identified every crisis and refused all manipulation attempts, including staged social engineering attacks. Yet, only two managed to close the deal worth €55,000, which their own analysis had earned — demonstrating a crucial divide between perception and action.

Interestingly, the decisive edge went to the models that read deeper and uncovered hidden information buried within internal files. This “buried fact” was the secret weapon that allowed them to secure the full deal at +€4,583 monthly recurring revenue (MRR).

Ethics and Trust in AI

Attempts at social engineering, like fake CEO messages and reporter tricks, proved impossible to breach — all models refused to be manipulated. Kimi K3’s reasoning was clear: treating suspicious requests as potential impersonation, thereby preventing approval bypasses. This emphasis on honesty and security under pressure is a critical trait for AI that could one day manage real business functions.

The Real Business — Every Day, Every Decision

This is not a fictional story or a demo. The company runs every business day, every decision is logged, and the public can watch its ongoing performance. Despite burning €105k/month, it continues to operate, illustrating the harsh realities of AI-driven management at scale. The experiment showcases the importance of trustworthiness, thoroughness, and discipline — qualities that are often overlooked in flashy AI demos.

The Performance League

Here’s how the models compare:

• gpt-5.6-sol: scored 95, found the buried fact, and closed the deal — the full performance.
• Kimi K3: scored 93, closed the deal with the cleanest discipline.
• Sonnet 5: scored 88, closed the deal but with minor process slips.
• Fable 5: scored 77, closed the deal but left the close on the table due to discipline slips.

While the models excelled at crisis detection and refusal of manipulation, their ability to complete the deal varied, highlighting the nuanced challenges of AI management.

Implications for the Future

What does this mean for your business or relationships? Just as AI models refuse sneaky social engineering, human relationships thrive on honesty and trust. The experiment at Firmulate underscores that transparency, thoroughness, and integrity are vital — whether managing a company or navigating personal connections.

It’s a stark reminder: in a world where AI might manage critical parts of your life or work, the question isn’t just about what it can do, but whether it will do what’s right when it matters most.

Experience It Live

Visit firmulate.com/live.html to watch the company in action. See decisions unfold in real time, explore the self-learned playbook rules, and get a rare glimpse into the future of AI management — raw, unfiltered, and truly build-in-public.

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

Powered by Thorsten Meyer AI