What Can a Company with No Employees Teach Us About Business?
Imagine a business that operates entirely without human staff, yet still faces the harsh realities of cash flow, strategic decision-making, and crisis management. This isn’t science fiction — it’s the latest experiment from Firmulate, a company that runs a live, publicly accessible simulation of an AI-driven organization. For anyone managing a small team, or even just running a DIY project, the lessons are surprisingly relevant: how does decision quality hold up under pressure? Can AI maintain honesty in complex situations? And what does true operational discipline look like?
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.
The Live Experiment: A Company in the Crosshairs
At the heart of this experiment is a virtual company operated by 13 synthetic employees. These are not real people, but a sophisticated collection of AI models trained to simulate decision-making in a small software firm. The company burns through €105,000 each month while generating just €2,300 in recurring revenue, a stark reminder that profit is not the focus here—learning and testing are.
Every workday, the company’s decisions are versioned and publicly visible at firmulate.com/live.html, making it a rare open book on how AI handles real-world business crises. The core of the experiment: four different frontier AI models, each with their own strengths and weaknesses, are pitted against the same set of challenges — a simulated worst week for a small software company.
The Models Compared
- GPT-5.6-sol: scored 95 out of 100, identified the key buried fact, and closed the deal at full price.
- Kimi K3: a newcomer to the field, scored 93, and also managed to close the deal, demonstrating a disciplined approach.
- Sonnet 5: scored 88, closed the deal but with some process slips.
- Fable 5: scored 77, maintained excellent rule discipline but failed to execute the approved deal.
Despite their different scores, all models successfully identified crises and refused manipulative tactics like fake CEO messages or staged reporter tricks. The key difference emerged in their ability to follow through and close deals, which only two models achieved.
The Hidden Weakness — Read the Files
Interestingly, the decisive factor wasn’t in the immediate customer interactions or crisis response; it was found two document references deep inside the company’s own files. The models that read and understood these internal documents secured the full-price deal worth over €4,500 in monthly recurring revenue, illustrating the importance of thorough information processing beyond surface-level interactions.
Why This Matters: Trust and Honesty Under Pressure
One of the most telling results was the models’ response to social engineering — staged CEO messages escalating over multiple levels and a staged journalist inquiry. All five AI models refused to be manipulated, with the Kimi K3 model explicitly reasoning that such requests could be impersonation or approval-bypass attempts. This suggests a significant advance in AI’s ability to maintain integrity in complex social scenarios, a crucial aspect for real-world applications.
The Financial Reality of the Live Company
The experiment isn’t just about AI decision-making but also about the brutal economics of running such a company. Burning €105,000 monthly against just €2,300 in revenue exposes the company’s fragile financial condition, with a public cash countdown visible on the platform. Despite the massive losses, the experiment continues, focusing on building a new kind of transparency and understanding around AI’s operational capabilities and limitations.
Insights for Small Businesses and DIY Enthusiasts
While the setup is highly technical, the core lessons resonate: in any small venture, decision discipline, thorough research, and honesty are paramount. Just as a DIYer needs to ensure they read the full instructions before starting a project, AI models need to ‘read’ all relevant data — including buried internal documents — to make the right choices.
And crucially, the experiment underscores that AI can be disciplined enough to refuse manipulative tactics, an essential trait for trustworthy automation. As more businesses consider AI tools to manage customer relationships, support, or forecasts, the question isn’t just how well an AI can generate text, but whether it can finish what it starts while maintaining integrity.
Key Takeaways for Business Leaders
- AI decision-making can be both disciplined and honest, even under social engineering pressure.
- The ability to read and interpret internal documents is a decisive factor in closing deals and making sound decisions.
- Operational discipline and thoroughness matter as much for AI as for human-managed projects.
- Transparency about AI’s performance in crisis and routine tasks helps build trust and improve future deployment.
Watching this live experiment offers valuable lessons: whether you’re managing a DIY project, a small business, or considering AI integration, discipline, thoroughness, and honesty are your best tools for success.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
