Imagine a company with no human employees, losing €105,000 every month, yet still competing in a high-stakes business simulation — and you can watch it unfold live. This isn’t sci-fi; it’s the latest experiment in AI management, where models run the entire operation in real time, making decisions, facing crises, and even trying to close deals.
The Public Experiment in AI Management
At the heart of this groundbreaking project is Firmulate, an open, observe-as-you-go company emulator that runs an actual small software business with a twist: every decision, crisis, and move is determined by advanced AI models. The site, firmulate.com/live.html, lets anyone watch this digital company in action, day after day, as it grapples with challenges that would make many human managers sweat.
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.
The Setup: A Small Business Facing Big Challenges
In this experiment, four frontier AI models—each with varying capabilities—are tasked with managing the same company during its most difficult week. They deal with the same clients, the same crises, and the same temptations to cut corners or manipulate data. The goal? To see which model can best steer the company to success, measured by a simple but critical metric: closing a €55,000 deal.
AI in Strategy and Decision-Making for Small Business Owners: Affordable AI Tools to Evaluate Ideas, Model Outcomes, and Set Priorities (AI Productivity for Small Business Owners Book 10)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Results That Speak Volumes
- The winner: gpt-5.6-sol with a score of 95, identified a hidden detail buried in company files, and successfully closed the deal — demonstrating complete performance.
- Close behind: Kimi K3, the new entrant from Moonshot, scored 93, also closing the deal with the cleanest discipline among all models.
- Followed by: Sonnet 5, with a score of 88, managed to close but showed a few process slips.
- The laggard: Fable 5, scoring 77, exhibited strong rule discipline but failed to follow through on the signed deal, leaving money on the table.
XTOOL AD20 Pro OBD2 Scanner – No Subscription, Full System Car Diagnostic Scan Tool with AI Analysis, Wireless OBD Car Code Reader, Oil Reset, Performance Test, Voltage Test
【NO Subscriptions & Wide Vehicle Support】 AD20PRO obd2 scanner diagnostic tool is built for simple, long-term ownership with…
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Hidden Weakness: The Secret to Winning
While all models spotted crises and refused manipulation attempts—such as fake CEO messages—one crucial advantage determined the winner. Both top-performing models read beyond the surface, diving two document references deep into the company’s files. The model that did so found a key piece of information that others missed, enabling it to close the deal at a full €4,583 MRR (monthly recurring revenue).
The AI Teammate: A PRACTICAL GUIDE TO WORKING WITH AI IN SOFTWARE DEVELOPMENT
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Why This Matters for Business and Tech
For a public audience used to hearing about AI’s capabilities in chatbots and support tools, this experiment shifts the focus to something more critical: integrity, thoroughness, and the ability to deliver results under pressure. It’s not just about AI sounding convincing—it’s about whether these models can actually finish what they start, read your files comprehensively, and stay honest when stakes are high.
Building in Public: Transparency and Learning
This is not a black box. Every decision, every rule learned, and every version of the AI is openly documented and auditable. The model’s behavior can be tracked daily, providing a transparent view into how AI can be tested in complex, real-world-like scenarios. The site even offers a quiz that challenges management decisions based on the AI’s actions, emphasizing the importance of understanding AI’s true capabilities.
The Larger Implication: AI as a Workforce Tester
What if your next employee wasn’t human but an AI that you could test against your company’s worst week—without risking your actual business? Firmulate offers just that: a way for enterprises to run their own management wargames, in a safe, read-only environment, to gauge how AI might perform before deploying it into real operations.
The Reality of a Money-Losing Company
Despite the AI’s impressive decision-making, the simulated company is losing €105,000 every month against a modest €2,300 MRR. It’s a stark reminder that even the smartest AI cannot fix fundamental business economics overnight. However, the experiment underscores that AI can excel in decision clarity—spotting hidden facts, resisting manipulation, and closing deals—if guided correctly.
This public experiment with AI-run companies reveals that success depends not just on AI’s ability to mimic human conversation but on its capacity to thoroughly analyze data, resist manipulation, and reliably finish tasks. Watching these models fight for survival in real time offers valuable insights into the future of AI-managed workforces and decision-making systems.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
