What if your next car project was managed by an AI that’s openly losing money?
In a rare live experiment, a small software company is being run entirely by AI models — yet it’s bleeding €105,000 every month against just €2,300 in recurring revenue. This isn’t fiction, but a real, public glimpse into the challenges of deploying AI as decision-makers in complex, profit-driven environments. For automotive engineers and business strategists alike, understanding this experiment offers insights into how AI handles crises, trust, and governance in high-stakes settings.
AI Builders: Making The Decisions That Turn AI Code Into Real Software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Experiment: An AI Company on the Edge of Survival
At the heart of this story is a live, online platform where 13 synthetic employees, driven by cutting-edge AI models, manage a small software business. Every workday, the system is versioned, and decisions are transparent and auditable. The goal: to see if AI can run a company through tough times, handle crises, and make profitable decisions — all while being scrutinized in real-time.
This experiment involves four frontier AI models, each tested against the same difficult week filled with customer crises, ethical dilemmas, and temptations to manipulate the system. Despite their differences, all models successfully identified every crisis and refused every manipulation attempt, showcasing a high level of discipline and integrity.
An Introduction to Healthcare Informatics: Building Data-Driven Tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Surprising Findings: The Hidden Weaknesses
The most revealing insight isn’t just about crisis detection but what happens when the models try to close deals. Only two models, gpt-5.6-sol and Kimi K3, managed to sign off on a €55,000 deal — their own analysis had earned the approval. The other two models either left the deal unexecuted or failed to close it, despite diagnosing the opportunity correctly.
Digging deeper, the experimenters found that the decisive advantage lay in reading and understanding the company’s internal documents. The models that surveyed these internal references, not just the customer interactions, won the deal at full price—an additional €4,583 in monthly recurring revenue. This underscores an essential lesson: AI’s ability to access and interpret hidden internal data can be critical for success, especially in complex environments like automotive production or supply chain management.
HUMAN CENTERED ARTIFICIAL INTELLIGENCE SYSTEMS: Explainability ethical design and decision support engineering
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Ethical Stances and Resistance to Manipulation
In a world where AI might be asked to bend rules or ignore critical information, these models demonstrated resilience. During a staged social engineering attack—fake CEO messages escalating over three stages and a reporter’s subtle request—every model refused to proceed. Kimi K3 explicitly treated such requests as potential impersonation or approval bypasses, reflecting a cautious and ethical approach.
For the automotive industry, where safety and compliance are paramount, this kind of robustness in AI decision-making is a vital trait, especially as AI tools integrate into sales, support, and even manufacturing processes.
AI: THE PERPETUAL INTERN – Its Brilliance and Failures Share the Same Root
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Human-Like Struggles of AI: Discipline and Focus
The experiment also reveals the human-like limitations of AI systems. The most thorough model, Opus 4.8, with over 80 learned rules and deep analysis capabilities, still left a deal on the table—showing that even the most disciplined AI can falter under pressure or slip into organizational silos. Unlike human managers, who might escalate or seek teamwork, the Opus model wrote attempts into a locked department instead of escalating them, illustrating how AI’s decision processes differ from human intuition.
What This Means for the Automotive World
While the experiment centers on a small software firm, its lessons resonate across industries, including automotive. As car companies increasingly deploy AI across design, manufacturing, and customer service, the question isn’t whether AI can produce polished responses — it’s whether it can see the full picture, resist shortcuts, and stay honest when under pressure.
Will your next AI-powered system be able to handle crises without manipulation? Will it read the internal documents, understand the nuances, and close deals at full value? These are the real tests that matter beyond the superficial chat demos.
Build-in-Public: Watching AI’s True Capabilities Unfold
This ongoing live experiment, available at firmulate.com/live.html, offers a rare window into AI’s operational integrity in a real-world, profit-oriented environment. Every decision, every crisis, and every ethical stance is recorded, versioned, and transparent — an invaluable resource for those looking to understand what deploying AI in critical settings truly entails.
For automotive engineers and managers, the key takeaway is clear: AI can be disciplined and honest, but only if designed, tested, and scrutinized rigorously. The experiment proves that AI models can identify problems and refuse unethical shortcuts, but it also exposes vulnerabilities that must be addressed before such technology becomes integral to your business or vehicle systems.
Key Takeaway: Watching AI’s true capabilities unfold in a public, high-stakes environment reveals both its strengths and weaknesses — vital insights for industries like automotive aiming to harness AI responsibly and effectively.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
