Watching an AI-Run Software Company Fight for Survival in Real Time

Imagine a company with no human employees, losing over €100,000 every month, yet still fighting to close a €55,000 deal. This is not science fiction but a real-time experiment revealing the true potential—and limits—of AI management systems. For those invested in smart home tech and automation, it’s a glimpse into how AI could one day oversee complex operations with minimal human oversight.

The Experiment: An AI Company Under Pressure

Firmulate, a pioneering platform, hosts a live experiment where four advanced AI models run a small software company through its toughest week. This isn’t a demo or a scripted scenario—it’s a real management simulation, with actual money mechanics, public visibility, and every decision meticulously versioned and auditable. The company has 13 synthetic employees, burning €105,000 monthly against a mere €2,300 in monthly recurring revenue (MRR). Every workday, the models navigate crises, customer interactions, and ethical dilemmas, all in real-time.

Key Findings: AI as a Trustworthy Manager

All four AI models successfully identified every crisis and refused every manipulation attempt, including social engineering tactics like fake CEO messages and reporter tricks. Notably, only two of the models managed to close the €55,000 deal their own analysis had earned—a crucial indicator of their decision-making integrity and discipline.

The decisive advantage lay in a buried fact—an internal document deep within the company’s files—that the models which read and understood this information secured the full deal at a full price, adding over €4,583 in monthly recurring revenue potential.

The Discrepancy in Performance and Discipline

The most thorough participant, Opus 4.8, with over 80 learned rules and deep analyses, finished last in the deal closing. It left a significant opportunity on the table and showed lapses in escalation discipline. Interestingly, all models faced the same weaknesses—indicating that even the most advanced AI struggles with certain process slip-ups under stress. Kimi K3, running without effort parameters, demonstrated the cleanest discipline and successfully closed the deal, highlighting how model configurations influence outcomes.

Implications for Business Automation

This experiment exposes a critical insight for companies looking to automate decision-making and management: success isn’t just about how well an AI writes or responds in chat. It’s about whether it can finish what it starts, understand the context deeply, and stay honest when under pressure. For smart home ecosystems and automation, this signals that AI’s role could soon extend beyond simple tasks to managing complex operations reliably and ethically.

Live and Transparent: Watch It Unfold

What makes this experiment unique is its transparency. The entire scenario runs live at firmulate.com/live. Viewers can see the company in action every business day, watch decisions unfold, read actual employee comments, and even explore the detailed logs of every AI decision—highlighting what works and where improvements are needed in real-time.

The Big Question: Are Your AI Tools Ready?

For owners of smart homes and automation systems, the message is clear: the question isn’t whether AI can produce convincing chat messages. It’s whether these AI agents can deliver consistent, honest, and complete work that aligns with real business goals. As AI models improve, their ability to navigate crises, read critical internal documents, and stay disciplined under pressure will determine their true value in managing the complex systems you rely on daily.

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

Powered by Thorsten Meyer AI