Watch AI Manage a Live Business Battle in Real Time — and Decide Its Fate

Imagine watching an AI-run company navigate crises, convince clients, and make tough decisions — all in real time. This isn’t a sci-fi scenario; it’s happening now, live, on the web. For creators and tech enthusiasts, it offers a peek into the future of AI-driven management and decision-making that could reshape how businesses operate, support, and innovate.

The Live Company Experiment: Reinventing Business Management with AI

At the forefront of artificial intelligence innovation, a public experiment called Firmulate is turning the idea of AI management into a visible, watchable spectacle. The setup is simple: a small, virtual software company with 13 synthetic employees faces real crises, real money mechanics, and real temptations. Every day, the company’s decisions are made, versioned, and displayed for the world to see. This isn’t a demonstration in a closed lab; it’s a live, ongoing battle testing how AI models handle management challenges under pressure.

The company operates with a clear financial backdrop: it burns €105,000 every month but earns just €2,300 in monthly recurring revenue. It’s running against a public cash countdown, making every decision critical, every mistake costly. The goal is to see if AI models can run this tightrope walk, make honest decisions, and successfully close deals, all while navigating crises with integrity.

The Models at Play: Different Approaches, Diverging Results

Four advanced AI models, or frontier models, each faced the same grueling test: run the company through its worst week, with identical customers, crises, and temptations. They were evaluated on their ability to detect crises, refuse manipulative tactics, and close a crucial €55,000 deal that the company’s own analysis had identified as the right choice.

The results were revealing:

• All four models identified every crisis and refused every attempt at manipulation, demonstrating strong ethical adherence.
• Only two models were able to close the deal at full price, despite all of them diagnosing the potential in the same way and delivering the same sales pitch.
• The decisive factor was a buried fact in the company’s internal files—only the models that read and analyzed these documents thoroughly were able to secure the contract at full value, worth an additional €4,583 in monthly recurring revenue.

Trust, Discipline, and Hidden Vulnerabilities

Beyond the obvious success or failure to close deals, the experiment uncovered deeper insights. For instance, one model, Opus 4.8, was the most meticulous, analyzing over 80 learned rules and providing the deepest insights. Yet, it finished last in the final outcome because it left the close on the table, showing discipline lapses—such as writing attempts into a restricted department instead of escalating them properly.

Another interesting facet was the AI models’ response to social engineering tactics. A staged fake CEO message escalating over three stages and a reporter trick asking for a quick ‘yes/no’ response were both refused by all models. Kimi K3 explained its refusal as a suspicion of impersonation, highlighting that these models are programmed to treat such requests as potential security threats.

The Bigger Picture: What This Means for Business and Creators

This experiment’s significance extends beyond pure tech curiosity. It raises foundational questions for any organization considering AI automation:

• Will your AI agents be honest and trustworthy when under pressure?
• Can they read and understand complex internal documents, or do they miss crucial hidden information?
• What does effective, ethical work cost in terms of resources and discipline?

For creators and tech innovators, this live showcase is a glimpse into a future where AI doesn’t just chat or generate content — it manages, strategizes, and possibly governs. Watching the ongoing experiment at firmulate.com/live offers a rare window into how autonomous AI companies could shape tomorrow’s business landscape.

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

Powered by Thorsten Meyer AI