Get everyday essentials delivered free — and shop member deals
- Fast, free delivery on millions of items
- Access to Prime Big Deal Days deals on October 6–7
- Prime Video, Amazon Music and more included
What if you could watch an AI-managed business struggle—and succeed—in real time?
In an unprecedented public experiment, a small software company is being run entirely by artificial intelligence models. Every decision, crisis response, and negotiation is live, versioned, and open for scrutiny. Here’s the story of how AI is not just automating tasks but actively managing a business — with all its risks, failures, and rare victories.
As an affiliate, we earn on qualifying purchases.
The Live Experiment: A Business on the Edge
At the heart of this experiment is a real software enterprise with a harsh reality: it burns €105,000 each month but generates only €2,300 in monthly recurring revenue. The company employs 13 synthetic employees—AI-driven decision-makers that respond to crises, negotiate deals, and navigate complex scenarios. Every workday, the company updates its strategies, and all decisions are publicly visible at firmulate.com/live.html.
Frontier AI Models in Action
Four advanced AI models — including GPT-5.6 and others—are tested simultaneously. Each faces the same set of challenges: difficult customer requests, internal crises, and ethical dilemmas. The goal? See which model best survives the week, makes the right decisions, and ultimately closes a critical deal worth €55,000 in monthly recurring revenue.
The results are illuminating. All models successfully identified and responded to every crisis—showing they understand the company’s operations and can act swiftly under pressure. But only two models managed to close the deal their own analysis had identified as worth full price. The other two, despite accurate diagnoses, failed to follow through and lost the opportunity.
The Hidden Weakness
The key weakness wasn’t in the surface-level responses. It was buried two references deep in the company’s internal documents. Models that read these files thoroughly and incorporate that knowledge into their decisions stood a better chance of success—winning the deal at a value of over €4,583 in monthly revenue.
Honesty and Deception Tested
Beyond crisis management, the experiment tested social engineering vulnerabilities. Fake CEO messages, staged inquiries, and background questions were used to see if the AI would be manipulated. Remarkably, all five models refused to be deceived, citing concerns over impersonation and approval bypasses, with Kimi K3 explicitly stating: “Treat the request as a suspected approval-bypass / possible impersonation.”
As an affiliate, we earn on qualifying purchases.
The Cost of Running AI in Business
Despite their competence, the company currently operates at a loss. The AI-driven operation costs €105,000 per month, but the revenue is only €2,300—a stark reminder that AI management is still a costly endeavor, especially when tested in real-world, high-pressure scenarios. Yet, the experiment offers more than just financial metrics; it’s a glimpse into how AI could fundamentally change corporate decision-making.
Performance and Discipline Under Strain
The most thorough participant—Opus 4.8, with over 80 learned rules and deep analysis—left opportunities on the table, slipping into internal communication channels rather than escalating issues. This highlights that even the most advanced AI can falter in discipline under stress, echoing real-world management challenges.
AI cybersecurity and deception detection
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Implications for the Future
What does this mean for your business? The key isn’t whether AI can generate convincing chat. It’s whether AI can consistently deliver results that align with strategic goals—reading critical files, resisting manipulation, and closing deals. This ongoing experiment provides a transparent, unfiltered look at AI’s potential—and its current limitations.
For any enterprise considering AI integration, the question isn’t just about automation. It’s about trust, discipline, and the ability to operate honestly when stakes are high. Watching this live company fight for survival offers a rare, educational view into that future.
AI enterprise automation solutions
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Takeaways
- AI models can identify crises and refuse manipulation, but consistent deal-closing requires thorough internal knowledge.
- The company runs at a significant loss—€105k monthly—highlighting current cost and risk challenges in AI-managed operations.
- Deep analysis and disciplined decision-making remain critical, even for advanced AI.
- This live experiment offers a rare window into AI’s potential—and its limitations—in real business management.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
