Watch AI Companies Fight for Survival in Public: The Extreme Experiment That Reveals True Business Readiness

Imagine running a business where every decision is scrutinized, every crisis tested, and the entire operation is publicly visible—yet you have no employees, and you’re losing money daily. This is the radical world of Firmulate, a live experiment that turns AI models into full-fledged, functioning companies. For garden and outdoor living enthusiasts, it might seem distant, but beneath this experiment lies a crucial lesson: the true test of AI’s usefulness isn’t just in generating convincing chat or ideas. It’s about reliability, honesty, and the ability to deliver real results under pressure.

The Public Company Without Employees

Firmulate’s live setup is unlike any traditional business. It’s a simulated software company running with 13 synthetic employees—each an AI model—whose activity can be watched in real-time at firmulate.com/live.html. The company is burning through €105,000 every month while generating only €2,300 in monthly recurring revenue. Its cash countdown is public, and every decision it makes is logged and versioned daily, providing an unprecedented window into its operational challenges.

Testing AI in the Trenches

This experiment pits four frontier AI models—each with different capabilities—against the same series of crises and scenarios. The goal? See if they can manage a small software firm through its toughest week. Every crisis, from customer support issues to internal crises, is identical across models. The models’ decisions are transparent and auditable, allowing observers to understand how each one responds to real-world business pressures.

What Did the Models Achieve?

• All four models identified every crisis, demonstrating sharp crisis perception.
• All refused manipulation attempts, including social engineering tricks like fake CEO messages and reporter entrapment.
• Only two models, gpt-5.6-sol and Kimi K3, managed to sign the €55,000 deal they had identified as worth full price, based on their analysis.
• Interestingly, the decisive edge came from reading a document buried two levels deep in the company’s internal files, an information that was not apparent in customer interactions but proved critical to closing the deal at full price.

The Reality of a Digital Business

This company’s daily operations illustrate the harsh reality of running an AI-powered business: despite advanced analysis and crisis management, it still struggles with execution discipline. For instance, the most meticulous model, Opus 4.8, analyzed deeply but left the close on the table, slipping into a less disciplined approach and leaving potential revenue unrealized. This highlights an important truth: even the best AI models can falter in execution, not just in diagnosis.

What This Means for Your Business

While the context here is a virtual company, the lessons are clear. If AI is to touch your operations—be it customer support, supply chain, or sales—the crucial questions go beyond how convincingly it can write or speak. Instead, you should ask: Will it complete the task? Will it read the relevant internal data before acting? Will it remain honest under pressure? And importantly, what is the cost of a unit of useful, trustworthy work?

Insights from the Live Experiment

The live leaderboard shows gpt-5.6-sol leading with a score of 95, followed closely by Kimi K3 at 93, then Sonnet 5 at 88, and Fable 5 at 77. Notably, K3 ran without an effort parameter, making its discipline even more impressive. The results underscore that even small differences in model behavior can significantly impact outcomes, especially in high-stakes business scenarios.

The Bigger Picture

This experiment, accessible at firmulate.com/quotes.html, is more than a demonstration; it’s a warning. AI models can be tested in a controlled, transparent environment—much like a wargame—to evaluate their readiness for real-world business challenges. For companies considering AI integration, it’s a reminder that trustworthiness, reliability, and discipline are critical metrics that go beyond surface-level performance.

How to Engage with the Experiment

You can observe the ongoing live runs at firmulate.com/live.html, or try running your own business simulation using their tools. This ‘run before you hire’ approach helps identify weak spots before deploying AI in critical roles. It’s a practical step toward ensuring that AI investments translate into dependable, honest, and effective business operations.

Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html

Powered by Thorsten Meyer AI