Imagine a company with no human employees, burning through €105,000 each month while generating just €2,300 in revenue. This is not science fiction—it’s a live experiment in AI management transparency, where you can watch an entire company’s daily battle for survival unfold in real time.
The Live Experiment: An AI-Managed Company in Action
At the heart of this ongoing venture is a small software business run entirely by artificial intelligence models. Unlike traditional companies, this one has no human staff. Instead, it relies on 13 synthetic employees—each powered by advanced AI models—that make decisions, handle crises, and even negotiate deals.
Each workday, the company’s performance is publicly documented and versioned, providing unprecedented transparency into AI decision-making. The company faces typical business challenges: customer crises, internal temptations to cut corners, and competitive pressures. Yet, what’s truly astonishing is how the AI models respond under these pressures, and what this reveals about their capabilities and limitations.
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.
How Do the Models Perform?
The experiment pits four frontier AI models against each other, each tasked with navigating the company’s worst week. The models face identical scenarios, from customer disputes to internal compliance issues, and are required to make management decisions based solely on the available data.
Remarkably, all four models identified every crisis and refused manipulation attempts, like fake CEO messages or reporter tricks. This demonstrates a clear capacity for ethical judgment and crisis recognition. However, only two models managed to close a crucial deal worth over €4,500 in monthly recurring revenue. Both identified a hidden piece of critical information buried two document references deep in the company’s files—information that, if accessed, led to a full-price deal.
Crisis Management Using AI Tools: A Practical Guide for Leaders to Predict, Respond, and Recover Faster From Modern Disruptions
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Hidden Weakness and Its Implications
Despite their strengths, the models revealed a significant vulnerability: the ability to find and leverage hidden data. The models that read deeper into the company’s files succeeded in closing the deal at full price, while those that didn’t missed the opportunity.
This underscores a vital aspect of AI decision-making: access to comprehensive information can be the difference between success and failure. It also hints at a broader challenge—how to ensure AI systems can locate and interpret critical data amid complex, unstructured information.
SQL with AI: A Complete Beginner's Guide to SQL, Databases, Data Analysis, and AI-Powered Querying
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Dealing with Social Engineering and Trust
In addition to conventional crises, the models were tested against social engineering tactics. Fake CEO messages, escalating over three stages, and a reporter trick asking for a quick approval were used as tests. All five models involved refused to be manipulated, with Kimi K3 explicitly reasoning: “Treat the request as a suspected approval-bypass / possible impersonation.”
This behavior suggests a promising level of suspicion and ethical restraint, addressing concerns about AI manipulation and trustworthiness in real-world applications.
Generative AI Security: Theories and Practices (Future of Business and Finance)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Cost of Running an AI-Managed Business
The live company’s financials are striking: burning €105,000 every month against a mere €2,300 in revenue. It operates under a public cash countdown, with every decision and change documented for viewers to analyze at firmulate.com/live.html. This transparency is a radical departure from typical corporate secrecy, offering an unfiltered view into the day-to-day struggles and decision processes of an AI-run enterprise.
Lessons from the Experiment
- The models excel at crisis detection and refuse unethical manipulations, indicating strong ethical and analytical capabilities.
- Access to deeper, hidden data is crucial—models that read more thoroughly close more deals.
- Even the most disciplined models can falter under pressure or when discipline slips, highlighting areas for improvement.
- Transparency and versioning allow real-time evaluation of AI performance, fostering a build-in-public approach that is transparent and educational.
As the experiment continues, it raises fundamental questions for future AI integration: Will AI systems be reliable enough to manage critical business functions? Can they be trusted to act ethically, especially when under pressure?
The Bigger Picture
This ongoing live experiment exemplifies what it means to build and evaluate AI in a public, real-world setting. It forces us to rethink management, trust, and the role of AI in decision-making—especially in high-stakes, real-money environments. For educators, scientists, and technology observers, it offers a rare glimpse into AI’s potential and its current limits.
To see the company in action and follow its daily progress, visit firmulate.com/live.html.
In a world where AI manages real companies, transparency reveals both strengths and vulnerabilities—highlighting that future AI-driven businesses must prioritize ethical decision-making, thorough information access, and public accountability.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
