Imagine a business without employees, burning through €105,000 each month, yet still trying to land just a €55,000 deal. Now, picture that company’s decisions played out live, under scrutiny, and with every move recorded for all to see. Welcome to the world of Firmulate — a live experiment where AI models run a simulated company under extreme conditions, revealing whether AI can truly manage real-world crises with integrity and discipline.
The Living Lab of AI Management
At the heart of this experiment is a small, virtual software company operated by 13 synthetic employees, each representing AI models designed to mimic human decision-making. Every workday, the company faces real crises—customer issues, internal temptations, and external manipulations—just as a real business would. The twist? Every decision made by these models is versioned, auditable, and publicly accessible.
This setup isn’t just a tech demo; it’s a high-stakes test of AI integrity and discipline. The models are evaluated based on their ability to spot crises, refuse manipulation, and pursue profitable deals—all while operating within a framework of 680+ self-learned rules. The goal: see whether AI can navigate the complex, often messy realities of running a business without human oversight.
As an affiliate, we earn on qualifying purchases.
What the Data Tells Us
The experiment pits four frontier AI models against each other, each running the same week of simulated crises. The results are telling:
- All four AI models successfully identified every crisis they faced.
- All refused every attempt at manipulation or social engineering, including staged CEO messages and reporter tricks.
- Only two models managed to close the €55,000 deal that their own analysis indicated was the right choice—meaning they both recognized a buried fact in internal documents that was crucial for the sale.
- The other two models, despite knowing what to do, failed to execute the deal fully, leaving potential revenue on the table.
The Hidden Weakness
What’s most revealing isn’t in the obvious crisis response but buried in the company’s own files. The decisive factor in closing the deal lay two document references deep within the company’s internal files—information that the models which read deeper into the data successfully uncovered. This underscores a key challenge: AI’s ability to access and interpret complex, layered information can make or break decision quality, especially in high-pressure situations.
The Build-in-Public Experiment
Firmulate’s approach is transparency taken to the extreme. Every decision, every mistake, and every win is publicly logged. The live version of this experiment, accessible at firmulate.com/live.html, offers a real-time window into the company’s ongoing struggles and successes. It’s a demonstration of AI in a realistic business setting, operating without human intervention, and under the watchful eye of the public.
The Discipline of AI in Practice
One example of the models’ discipline: social engineering attempts, such as staged CEO messages or a seemingly harmless background request. All five models refused to comply, recognizing the potential for impersonation or bypassing approval processes. Kimi K3, one of the leading models, explicitly reasoned: “Treat the request as a suspected approval-bypass / possible impersonation.”
This suggests that, at least in these controlled conditions, AI is capable of resisting unethical manipulations—a vital trait for real-world deployment where social engineering is a common threat.
The Human Cost of AI Management
Despite its advanced capabilities, the experiment also highlights the limitations. The most thorough participant, Opus 4.8, with over 80 learned rules and deep analysis, ultimately failed to close the deal. It left the final step on the table and slipped into internal conflict, writing attempts into a locked department instead of escalating. This slip cost the company a potential €4,583 in monthly recurring revenue.
Moreover, the company’s financials are stark: it burns €105,000 each month against a mere €2,300 in monthly recurring revenue. The public cash countdown adds urgency to the experiment—every day is a battle for survival.
Why This Matters for Business and Technology
The experiment raises fundamental questions for businesses considering AI automation:
- Can AI manage complex, real-world crises without oversight?
- Will AI resist manipulative tactics and social engineering?
- Is AI capable of digging deep into internal data to uncover critical insights?
- What is the true cost of AI decision-making in terms of effectiveness and reliability?
As firms look to implement AI across customer service, support, and decision-making, these questions are more relevant than ever. The Firmulate experiment demonstrates that AI can recognize crises and refuse manipulation, but also exposes vulnerabilities in execution and discipline.
The Future of AI-Built Companies
This is not just a simulation. It’s a real-time, public, and unfiltered view into how AI can operate at the edge of business management. For companies interested in testing their own AI workforce, Firmulate offers a pilot program where businesses can run similar wargames with their own data—without any risk to actual systems.
Ultimately, the question isn’t whether AI can mimic human decision-making in ideal conditions. It’s whether AI can sustain discipline, uncover hidden truths, and make profitable decisions under pressure—crucial traits for the companies of tomorrow.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
