Watch a Public AI-Run Company Struggle to Survive Day by Day

Can AI Save or Sink a Company? Watch a Live Experiment in Action

Imagine a company with no human employees, battling daily crises, making thousands of decisions—yet losing €105,000 each month—and all of it is happening live for anyone to see. This is not science fiction; it’s the real-time experiment conducted by Firmulate, a pioneering AI-driven business simulation that offers a brutally candid view into how artificial intelligence behaves in high-stakes corporate scenarios.

A Transparent Look at AI in Business

At the heart of this experiment is a small, real software company run entirely by AI models. It operates with 13 synthetic employees, managing real money mechanics—burning €105,000 each month against a modest €2,300 monthly recurring revenue. Every decision made by these models is publicly visible, versioned daily, and scrutinized by anyone interested.

What makes this experiment particularly compelling is its honesty. Unlike typical AI demos that showcase only polished chat interactions, this setup tests AI decision-making under real-world pressures. The models face crises, customer demands, and even social engineering attempts, like fake messages from a CEO or a reporter seeking confidential approval. Remarkably, all four tested models correctly identified every crisis and refused manipulation attempts. Yet, only two managed to close a critical deal worth €55,000—a small victory in a sea of losses.

The Hidden Weakness in the Company’s Records

Deep within the company’s own files lay the crucial piece of information that could have saved the day. The models that read and understood this buried document referenced it, and as a result, they secured the deal at full price—adding €4,583 in monthly recurring revenue. This demonstrates a vital point: access to the right information at the right time is often the deciding factor in real business outcomes.

Why Does This Matter to You?

For those managing or overseeing senior care and aging services, the lesson is clear: AI’s ability to handle complex, high-pressure decisions is still evolving. When AI touches your operations—be it in scheduling, support, or forecasts—the question is not only whether it can generate convincing chat responses. The real test is: does it follow through on commitments, read and understand critical internal documents, and resist manipulation under stress?

Imagine AI-enabled systems that could manage patient data, coordinate care, or even navigate regulatory crises with integrity. Today’s experiments show promising signs but also reveal vulnerabilities. The key takeaway is that AI’s capacity for honest, effective work depends heavily on access to complete information and a disciplined approach—something still being tested in live settings like this one.

The Company’s Ongoing Struggle

The live experiment’s company is not profitable. It burns through €105,000 monthly, with only a small €2,300 incoming. Every day, the AI models face the same crises, temptations, and decision points, all publicly documented. The experiment is a brutally honest lens into the potential and limitations of AI in real-world corporate decision-making.

One model—Opus 4.8—stood out for its thorough analysis, learning over 80 rules and performing deep diagnostics. Yet, it left a deal on the table by hesitating on escalation instead of closing. Interestingly, when all four models were tested at the default API effort level, they performed similarly, underscoring that process discipline and clear decision parameters matter just as much as raw intelligence.

What’s Next? The Future of AI in Business

This ongoing experiment is more than a curiosity. It’s a real-time test of whether AI can truly replace or augment human judgment in business operations. The models’ ability to identify crises, refuse manipulative tactics, and close deals at full price reveals both the promise and the current limits of AI’s decision-making.

If AI agents will engage with your critical systems—whether managing customer interactions, financial forecasts, or compliance—the pressing questions are: Will they complete what they start? Will they stay honest under pressure? And what is the true cost of a unit of useful work?

Explore Further

For those interested in testing AI decision-making against their own business scenarios, the platform offers a read-only export feature. It allows companies to run simulations without risking their actual systems or data. You can see how your AI might perform in your context, with full transparency and control. Discover more at firmulate.com/pilot.html.

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

Powered by Thorsten Meyer AI