What if your garage could run itself—without human employees—yet still faces daily financial chaos?
Imagine a small business where every decision is made by an AI model, navigating crises, customer demands, and ethical dilemmas, all while losing thousands of euros each month. Welcome to the world of the live-company experiment by Firmulate, where AI models run a real, functioning software company in front of your eyes—every day, every decision, every risk.
Building AI-Powered Products: The Essential Guide to AI and GenAI Product Management
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Live Experiment: An AI Company in Action
Firmulate has set up a unique, transparent test bed: a real company with 13 synthetic employees, each driven by advanced AI models. Every day, the company faces the same set of crises, customer interactions, and ethical tests, with decisions being logged and made public at firmulate.com/live.html. The goal: assess whether AI can truly handle the complexities of running a business, especially under pressure.
The AI Models Tested
- gpt-5.6-sol: scored 95 out of 100, identified critical hidden information, and successfully closed a €55,000 deal.
- Kimi K3: scored 93, demonstrating the cleanest discipline and sealing the same deal.
- Sonnet 5: scored 88, also closing the deal but with some process slips.
- Opus 4.8: scored 73, with the deepest analysis but ultimately left the opportunity on the table.
Despite their differences, all models showed they could recognize crises and refuse manipulation attempts—an essential trait for trustworthy automation. Yet, only two models successfully signed the deal that paid the company’s bills, revealing a critical gap between diagnosis and action.
Crucial Discoveries and Weaknesses
The most telling weakness wasn’t in customer interactions but in internal company documents. The models that examined deeper files uncovered a vital piece of information buried two references into their files—this hidden detail was key to winning the deal and avoiding a €105,000 monthly burn rate.
Ethical and Social Challenges
Firmulate also tests AI’s moral compass through social engineering. Fake CEO messages and manipulated scenarios were presented to the models, all of which refused to comply—displaying a robust understanding that certain manipulations are dangerous and unacceptable. Kimi K3, for instance, identified the request as a potential impersonation, refusing to act.
AI in Strategy and Decision-Making for Small Business Owners: Affordable AI Tools to Evaluate Ideas, Model Outcomes, and Set Priorities (AI Productivity for Small Business Owners Book 10)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Reality of a Money-Losing Business
This experiment isn’t just theoretical. The company operates in real-time, losing €105,000 every month against a revenue of only €2,300. It’s a public, transparent portrait of a business fighting for survival—an unrelenting test of whether AI can be trusted with real work, even when the financial stakes are high.
The Deep Dive: Why Did Some Models Fail to Finalize?
Opus 4.8, the most thorough model with over 80 learned rules, left opportunities unclaimed—failing to escalate issues or finalize deals, instead writing internal memos resulting in discipline slips. This highlights that depth of analysis alone isn’t enough without decisive execution and discipline.
The Significance for Industry
Today’s AI tools are often judged on their ability to generate convincing chat responses. But in the real world—like a garage or manufacturing plant—it is whether they can handle crises, stay honest under pressure, and complete useful work that truly matters. This experiment from Firmulate pushes the boundary, showing that AI can spot issues and refuse manipulation but still struggles with execution under complex, real-world scenarios.
AI ethical dilemma simulation software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What This Means for Your Business
For automotive and garage operations contemplating AI adoption, the message is clear: it’s not enough for AI to sound convincing. The true measure is whether it can deliver consistent, trustworthy results without supervision, especially in high-stakes situations. Watching this experiment unfold provides a rare glimpse into the future of autonomous decision-making—where AI competes with human judgment and the stakes are visible, pressing, and real.
Key Takeaway
This live experiment demonstrates that AI models can recognize crises and avoid manipulative tactics but still struggle to close deals and execute decisively in a high-pressure, money-losing environment. For automotive businesses exploring automation, the critical question isn’t just about AI’s conversational skills but its ability to complete valuable, trustworthy work under real-world stress.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
AI Automation for REAL ESTATE AGENTS: Transform Your Real Estate Business with AI, Automation Tools, and AI-Powered Lead Generation Systems
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
