Imagine a startup with no human employees, burning through €105,000 every month yet still managing to operate, learn, and adapt—publicly. This is not fiction but the reality of a pioneering experiment in AI management, where every decision is tracked, and every crisis is faced head-on. What does this tell us about the future of work, trust, and automation? Let’s dive into a live showcase of AI in the crucible of real business challenges.
The Live Company: A Public Experiment in AI Management
At the heart of this experiment is a fully automated company—13 synthetic employees working tirelessly behind the scenes. Every business day, the company faces real crises, customer decisions, and manipulations, all in a publicly viewable environment at firmulate.com/live. The mechanics are straightforward but intense: burning €105,000 monthly against a modest €2,300 monthly recurring revenue, with a public countdown showing how long it can survive without additional funding.
Construction Program Management – Decision Making and Optimization Techniques
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The AI Models in Action
Four frontier AI models—ranging from GPT-5.6-sol to Opus 4.8—were tested by running them through the same brutal week of business crises. Each model faced the same customers, the same crises, and the same temptations to cheat or manipulate. The results offer a revealing glimpse into AI’s capabilities and limitations in a management context.
Outstanding Performance and Surprising Weaknesses
- All four AI models correctly identified every crisis and refused every manipulation attempt, demonstrating a robust grasp of ethical boundaries and crisis recognition.
- Only two models successfully signed the €55,000 deal their own analysis had earned, closing the deal at full price. The other two, despite diagnosing correctly and pitching well, left deals on the table or failed to execute them.
- Interestingly, the decisive factor wasn’t in the customer interactions but buried deep in the company’s internal files. The models that read these files won the deal at full rate—adding more than €4,500 in monthly recurring revenue.
Crisis Management for Software Development and Knowledge Transfer (Smart Innovation, Systems and Technologies, 61)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Building in Public: Transparency and Trust Under Pressure
The experiment extends beyond decision-making: it also tests social engineering resilience. Fake CEO messages and staged reporter requests were used to see if the AI would accept questionable instructions. All five models refused these requests, with Kimi K3 explicitly treating them as suspicious. This demonstrates an ability to maintain integrity in the face of psychological pressure—a critical trait for automation in sensitive roles.
Power BI for Data Analytics: The Lazy Person's Guide to Automated Reporting: Build It Once, Never Touch It Again (Data Analytics for Beginners)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Ongoing Struggle of an Autonomous Business
Despite the advanced AI capabilities, the company remains in a fragile state. It’s burning €105,000 every month while generating just €2,300 in recurring revenue. The public cash countdown emphasizes the urgency of every decision, highlighting the narrow margin for error when humans are replaced by AI in operational roles. Every day is a battle for survival, and each decision is meticulously versioned and auditable, creating a transparent record of how AI manages real business risks.
AI and Ethics: Adopting AI Technology Without Losing Our Humanity
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Insights from the Results
The most thorough model, Opus 4.8, analyzed over 80 learned rules and provided deep insights. Yet, it still underperformed relative to simpler models, sometimes slipping in discipline and leaving deals unexecuted. Meanwhile, Kimi K3, which ran without an effort parameter, achieved the cleanest discipline and successfully closed deals, showing that sometimes less complex configurations can perform better in practice.
The Bigger Picture for Business and AI
This experiment is more than a technical showcase; it’s a living proof-of-concept for how AI can be tested and validated in real-world scenarios. For companies, the question isn’t just whether AI can produce good writing or chat responses, but whether it can truly finish tasks, read critical documents, and stay honest under pressure. These are the qualities that will determine the next wave of automation and AI-driven management.
Further Engagement and Observation
Interested in seeing this in action? You can watch the ongoing live experiment at firmulate.com/live. For insights into how AI models decide, explore the detailed quotes and scores at firmulate.com/quotes.html. And for a fun challenge, try your hand at the management quiz—feature your intuition against AI decision-making at firmulate.com/quiz.html.
This public AI management experiment reveals that while models can identify crises and refuse manipulation, success depends on nuanced decision-making and internal data access. It underscores the importance of transparency, discipline, and trust in AI-driven business operations—lessons vital for the future of automation.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
