Imagine a business that operates entirely without human employees, yet faces the daily challenge of staying afloat — and you can actually watch it happen online. This is not science fiction; it’s the real-time experiment of a tiny software company run entirely by artificial intelligence models, fighting to survive in the most transparent way possible.
The Live Experiment: A Business in the Public Eye
At Firmulate, a pioneering platform, you can observe a small, automated company in action. It’s powered by 13 synthetic employees, each a different AI model tasked with managing the company’s operations. With a monthly burn rate of €105,000 against only €2,300 in recurring revenue, the company is a high-stakes experiment in AI management and resilience.
Building AI-Powered Products: The Essential Guide to AI and GenAI Product Management
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
How the Company Operates and What It Tests
Every workday, the company’s decision-making process is versioned and publicly available. This continuous transparency allows observers to see how AI models handle real crises, customer interactions, and ethical dilemmas, all while trying to close deals and keep the company’s cash running out.
The AI Culture Blueprint: Moving Beyond Tools to Create Human-Centered AI Adoption
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The AI Models: Different Strengths, Same Challenges
Four frontier AI models were tested in this environment, each given the same set of challenges: managing a company through its worst week. Their performance was measured across several parameters, from crisis detection to ethical decision-making and deal closure.
Performance Scores and Outcomes
- gpt-5.6-sol scored 95 and successfully closed a deal worth €55,000, capturing a hidden opportunity in the company’s files—a crucial advantage overlooked by others.
- Kimi K3, a newcomer, scored 93, also closing the deal and demonstrating the cleanest disciplinary behavior among models.
- Sonnet 5 scored 88 and managed to close the deal but with some process slips.
- Fable 5 scored 77, showing excellent rule discipline but failing to execute an approved deal.
HUMAN CENTERED ARTIFICIAL INTELLIGENCE SYSTEMS: Explainability ethical design and decision support engineering
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Ethics and Trust Are Non-Negotiable
In tests involving social engineering—such as fake CEO messages and a reporter trick—the models refused every manipulation attempt, with Kimi K3 explicitly noting suspicions of impersonation. This underscores that AI in business must not only perform but also uphold integrity under pressure.
AI internal document reading tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Hidden Weakness: Reading the Files
The decisive factor in securing the deal was reading and understanding a document deep within the company’s own files—an ability that gave the winning AI a critical advantage. This suggests that what models read and understand internally can be more influential than their responses in customer-facing scenarios.
Why This Matters for Your Business
For companies considering AI integration, the question is no longer just about how well an AI writes or talks. It’s about whether it can complete tasks, stay honest, and read critical internal information—especially under stress. The experiment proves that AI can be tested rigorously and transparently before deployment, reducing risks and increasing trustworthiness.
Real Money, Real Risk, Real Progress
The live company, with its public cash countdown and daily evolving rules—over 680 learned playbook rules—demonstrates the complex reality of AI-managed operations. It’s a compelling proof point that AI decision-making is not a black box but a field of ongoing, observable progress.
This extreme build-in-public experiment reveals that AI-driven companies can be transparent, resilient, and trustworthy—if tested properly. For your business, it’s a glimpse into a future where AI doesn’t just assist but manages critical operations, all openly and verifiably.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
