Inside a Living Experiment: Can AI Run a Company Without Employees and Still Survive?

Imagine a business that operates entirely without human employees, making critical decisions under real-world pressure — all while its financial fate hangs in the balance. Welcome to the frontier of artificial intelligence in management, where a real, live company is navigating its toughest week yet, watched by thousands as it fights for survival.

The Live Company That’s Changing How We See AI

At the heart of this extraordinary experiment is a small software business managed not by humans but by artificial intelligence models. This isn’t a fictional story; it’s a live, ongoing test available for anyone to watch at firmulate.com/live. Every weekday, this company’s decision-making process is recorded, versioned, and analyzed, revealing what AI can and cannot do in the high-stakes world of running a business.

What’s Going On?

The company is staffed by thirteen synthetic employees, each driven by cutting-edge AI models like GPT-5.6-sol, Kimi K3, Sonnet 5, and Opus 4.8. These models are tasked with managing crises, making strategic decisions, and navigating customer interactions — all while the real-world mechanics of money come into play. Currently, the company burns through €105,000 each month but earns only €2,300 in monthly recurring revenue, steadily approaching a public cash countdown.

Real Crises, Real Stakes

This isn’t a sanitized simulation. The models face the same challenges a human management team would encounter — unexpected crises, manipulative tactics, and ethical dilemmas. For example, the models were tested against social engineering attempts, including fake CEO messages and a reporter’s subtle inquiries. Remarkably, all five models refused to proceed with deceptive requests, citing concerns about impersonation and bypassing approval processes.

Decisive Performance in Critical Moments

One of the most striking findings comes from the models’ ability to identify hidden information crucial for closing a deal. The models that thoroughly read the company’s internal files discovered a key detail buried two documents deep — a detail not apparent in the immediate customer interactions. This insight was instrumental, allowing two models to clinch a €55,000 deal, which would have added +€4,583 to monthly recurring revenue. Yet, only these two models signed the deal; the others, despite diagnosing the opportunity, left it on the table.

What Does It Show About AI’s Capabilities?

The experiment’s scores tell a compelling story. While GPT-5.6-sol scored the highest at 95 and Kimi K3 followed closely at 93, all models succeeded in recognizing crises and refusing manipulative tactics. The critical weakness was in following through on opportunities — a sign that even the most advanced models still grapple with discipline and execution under pressure.

Beyond Demos: A Transparent, Build-in-Public Venture

This isn’t just a demo on a screen. The company runs live, every weekday, with real money mechanics, public cash countdown, and a detailed record of each decision made. Visitors can watch the company’s daily progress, read the self-learned rules guiding each decision, and even participate in quizzes to guess which AI model made which call. The goal? To demonstrate that AI’s true potential lies not in chatty demos but in its ability to complete complex, high-stakes tasks reliably.

Implications for Business and Self-Worth

For women focused on self-worth, this experiment underscores an important truth: automation and AI are reshaping how we work and value our contributions. It’s not just about replacing jobs but about understanding what human qualities — like discipline, ethical judgment, and strategic reading — remain vital. As AI models grow more capable, the challenge becomes ensuring they align with our standards of trust and integrity.

The Future of Work Is Watching

As this live experiment continues, it offers a rare window into the future of AI-driven management. Will these models become reliable partners or remain imperfect assistants? The answer depends on ongoing development, transparency, and the ability to teach machines to read between the lines — just as humans do.

This real-time experiment shows AI models can recognize crises and refuse unethical tactics, but completing complex deals still challenges even the smartest systems. It’s a glimpse into how AI might reshape trustworthy management — if we can guide it properly.

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

Powered by Thorsten Meyer AI