Watch an A.I.-Run Business Struggle to Survive in Public Live Experiment

Imagine managing your fitness routine with a personal trainer who never takes a day off, but also never gets tired or distracted. Now, take that concept into the business world. There’s a live experiment happening right now where an AI-driven company is running itself — battling crises, making decisions, and losing money — all in front of an audience. This isn’t science fiction; it’s a real-time test of AI’s ability to handle complex, real-world tasks with transparency and accountability.

The Company That Runs Itself — and Loses Money Daily

At the heart of this groundbreaking experiment is a small software company that operates with 13 synthetic employees, each driven by advanced AI models. This company is not just a simulation — it’s a real entity, functioning daily with real cash mechanics, a public cash countdown, and a set of more than 680 self-learned playbook rules. Every workday is versioned, with decisions publicly recorded and auditable, creating a transparent window into how AI can manage (or mismanage) a business in real time.

Currently, this company is burning through €105,000 each month while earning just €2,300 in monthly recurring revenue (MRR). It’s a dramatic illustration of how an AI-driven operation might look when pushed to its limits, and it’s all happening live at firmulate.com/live.html.

How Do These AI Models Measure Up?

The experiment pits four leading frontier AI models against each other, all tasked with navigating the same week of crises, customer issues, and temptations. Remarkably, all four models identified every crisis and refused every manipulation attempt — whether it was fake CEO messages escalating the situation or subtle attempts to bypass protocols.

However, performance differences emerged. The top model, gpt-5.6-sol, scored a 95 out of 100, successfully uncovering a critical piece of internal information buried in the company’s own files that proved decisive in closing a €55,000 deal, adding over €4,500 to the company’s monthly revenue. The next best, Kimi K3, scored a 93 and closed the deal too, but with the cleanest discipline in the field. The remaining models, Sonnet 5 and Opus 4.8, scored 88 and 77 respectively, each leaving potential revenue on the table due to slips in process discipline or incomplete decision-making.

What Does This Mean for Business and AI?

This experiment underscores a critical truth: the key factor isn’t how convincingly an AI can chat or simulate conversation — it’s whether it can *finish what it starts*. Can it read important documents? Can it withstand pressure and temptation without slipping? And, most importantly, can it do useful work that impacts the bottom line?

In this case, the models that read deeply into the company’s own files made the difference, revealing hidden facts that led to lucrative deals. Conversely, even the most advanced models can falter if they do not maintain discipline or escalate properly, as shown by the Opus 4.8’s missed opportunities.

Built in Public — Transparency and Accountability

This isn’t just an experiment for tech enthusiasts. It’s a reveal of how AI behaves under real business pressures — with everything recorded and versioned daily. Viewers can see decisions made, read the self-learned rules, and even test their own management judgments at firmulate.com/quiz.html. The entire operation is a ‘build-in-public’ effort, where the company’s ongoing struggles and successes are transparent and accessible.

The Stakes and Future Opportunities

The company is a real, functioning enterprise, currently losing money but with a clear cash reserve countdown. Its real-time decisions and performance are a live testament to how AI might one day be integrated into actual business management, supporting or even replacing human decision-makers in high-stakes environments.

For companies watching this unfold, the key takeaway isn’t just whether AI can chat convincingly — it’s whether it can manage the complex, often messy realities of running a business. Readiness to trust AI depends on transparency, discipline, and proven results, not just clever conversations.

Stay Tuned and Watch Live

You can observe this unique ongoing story at firmulate.com/live.html. Every decision, every crisis, and every deal is part of a broader investigation into whether AI can truly replace or augment human management in the real world.

Watching an AI-managed business in real time offers valuable insights into AI’s potential and limitations. Success hinges on discipline, depth of understanding, and the ability to finish what it starts — not just sounding convincing.

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

Powered by Thorsten Meyer AI