Imagine a company that has no employees, loses €105,000 every month, yet continues to operate publicly — and you can watch it battle for survival day by day. This is not science fiction, but the reality of a live experiment demonstrating how AI models manage complex business decisions under pressure.
The Live Company That Keeps Running
At the heart of this experiment is a real, functioning software company — not a simulation. It employs 13 synthetic employees, each governed by a set of more than 680 self-learned playbook rules, with decisions made and versioned daily. Every workday, its actions are publicly accessible at firmulate.com/live.html. Here, viewers see a company fighting to stay afloat, burning €105,000 in cash monthly against a modest €2,300 monthly recurring revenue (MRR).
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.
Why Does It Matter to You?
For personal finance and investing enthusiasts, the core question is: how reliable are AI systems in managing critical tasks, especially when they have real money on the line? The experiment reveals that AI models can detect crises and resist manipulation attempts — crucial traits when AI is poised to touch your CRM, customer support, or forecasting systems. But the key is not just whether they can write convincingly; it’s whether they can finish what they start, stay honest under pressure, and make genuinely useful decisions.
Project Management Tools (AI for Risks)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Experiment: Testing AI Under Extreme Pressure
Four leading AI models — including GPT-5.6-SOL, Kimi K3, Sonnet 5, and Opus 4.8 — were tasked with navigating the same worst week of a small software business. With the same set of customers, crises, and temptations, each model’s decisions are meticulously recorded and auditable. The goal: see whether AI can navigate real-world challenges without cheating or slipping up.
AI internal document reading tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Surprising Results
- All four models identified every crisis and refused manipulation attempts, demonstrating integrity under pressure.
- Only two models managed to close a critical €55,000 deal — the same deal their own analysis had identified as worthwhile.
- The key weakness was hidden deep within the company’s internal files — a detail only the models that read these files could leverage to win the deal, adding €4,583 MRR.
- Attempts to engineer social engineering tricks, like fake CEO messages or reporter-style background questions, were refused by all models, with Kimi K3 explicitly reasoning about impersonation risks.
![Claude AI for Beginners Bible: [5 in 1] The Ultimate Guide to Automate Your Work, Save Hours Every Week, and Use AI for Real-World Results](https://m.media-amazon.com/images/I/415+fSJacsL._SL500_.jpg)
Claude AI for Beginners Bible: [5 in 1] The Ultimate Guide to Automate Your Work, Save Hours Every Week, and Use AI for Real-World Results
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Reality of the Live Company
This company isn’t just a demo; it’s a real operation, losing money month after month, with its fate visible at firmulate.com/live.html. It runs every business day, with decisions made by AI models that are constantly evolving and learning. Every decision is versioned, offering a transparent view into its decision-making process.
What the Results Say About AI in Business
The experiment emphasizes that success isn’t just about AI’s ability to generate text or chat convincingly. When AI manages real money and makes critical decisions, attributes like honesty, thoroughness, and attention to internal data become paramount. The models that read internal documents and resist manipulation are the ones that succeed in closing deals and making profitable decisions.
The Broader Implications
For investors and business leaders, this experiment offers a sobering reality check. AI systems are not infallible; they can identify crises and refuse to manipulate when properly designed. But the difference in performance often hinges on internal data access and decision discipline. As AI continues to integrate into operational systems, understanding these nuances becomes vital for managing risk and maximizing returns.
See It Yourself
Visit firmulate.com/live.html to watch the company in action, explore the decisions, and see which models succeed or falter. You can also engage with a quiz at firmulate.com/quiz.html to test your ability to guess which AI model made which decision.
Final Takeaway
As AI agents become more embedded in core business functions, their ability to reliably finish tasks, stay honest, and leverage internal data will determine their true value. The live experiment by Firmulate offers a rare, unvarnished view of AI’s potential and its current limits — a must-watch for anyone invested in the future of business technology.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
