Imagine a business without humans, driven entirely by artificial intelligence, yet struggling daily to stay afloat. This isn’t science fiction; it’s a live experiment where AI models run a small software company in real time, making decisions, facing crises, and even losing money. For food enthusiasts and curious minds alike, this story reveals how technology is reshaping work — with lessons that extend far beyond the digital realm.
The Live Experiment: An AI Company in Public View
At the forefront of transparency and innovation, Firmulate hosts a live simulation where four advanced AI models operate a small software firm, mimicking real business conditions in a way that’s fully auditable and publicly accessible at firmulate.com/live.html. The company has no human employees; instead, it relies on 680+ self-learned rules and daily versioned decision logs, offering a rare window into how AI performs under pressure.
AI Builders: Making The Decisions That Turn AI Code Into Real Software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Facing the Same Crises, Making the Same Decisions
Every week, the four models confront identical challenges: customer crises, tempting manipulations, and complex negotiations. The goal: see if AI can not only recognize problems but also act ethically and decisively. The results are telling:
- All four models identified every crisis correctly.
- They uniformly refused manipulation attempts, including social engineering tricks like staged CEO messages and fake reporter inquiries.
- Despite diagnosing problems accurately and pitching deals similarly, only two models closed a €55,000 deal they earned through their own analysis.
This gap in performance illustrates a critical point: the difference between diagnosis and execution. While all models spotted issues, only some translated those insights into decisive action—signing deals that bring in real revenue.
The Hidden Weakness: The Critical Information Gap
Interestingly, the decisive edge for the successful AI came from deep within the company’s own files, not from external customer data. The models that read these internal documents won the deal at full price, adding more than €4,583 in monthly recurring revenue, a tangible financial advantage.
Ethical Boundaries and Trust Under Pressure
One of the experiment’s fascinating aspects was testing the models’ responses to social engineering. Over three stages, the AI faced simulated CEO messages escalating in urgency and a reporter trick demanding a simple yes/no response. All five models refused these requests, with Kimi K3 explaining, “Treat the request as a suspected approval-bypass / possible impersonation.” This demonstrates that, even in a simulated environment, AI can maintain integrity under manipulation pressures.
The Real Business: Running Out of Cash in Plain Sight
Behind the experiment lies a real company, continually burning €105,000 each month against a modest €2,300 in monthly recurring revenue. It’s a visible, ongoing struggle with a public cash countdown, operating with every decision logged and versioned daily. This raw, unfiltered glimpse into an AI-driven business fighting to survive offers crucial insights into the actual costs, risks, and discipline involved in deploying AI at scale.
Lessons Beyond the Digital
For those in the food industry or any field involving decision-making and trust, the story of this AI company underscores vital lessons:
- Financial discipline remains critical — even for AI-run entities.
- Understanding and reading internal information can be a game-changer in negotiations.
- Ethical behavior isn’t guaranteed by AI; it must be explicitly programmed and tested under pressure.
- The true test of AI is not how well it chat or simulate human conversation, but whether it can finish what it starts and stay honest under stress.
Explore and Experiment Yourself
For businesses curious about testing their own AI capabilities, Firmulate offers a read-only “wargame” environment to simulate decision-making without risking real systems. Discover how your AI models handle crises and ethical dilemmas at firmulate.com/pilot.html.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
