Imagine a business without humans, where AI models are the only decision-makers, and every move is watched by the world. This is not science fiction, but a live experiment demonstrating how AI can manage a company under extreme pressure. For families and parents, it’s a glimpse into a future where automation and AI are not just tools, but active participants in real-world decision-making—shaping the products, services, and systems that touch everyday life.
The Live Company Running in Public
At the heart of this experiment is Firmulate, a company that operates completely transparently, with 13 synthetic employees making decisions about a small software business. Every workday, the company’s rules are versioned, and its performance is publicly available on the live site. What makes this so fascinating is that the company is losing €105,000 each month — more money than it’s earning in revenue, which stands at just €2,300 monthly.
This stark imbalance underscores that the experiment isn’t about profit but about testing AI models under conditions of real crisis and temptation. These models are tasked with managing the company through simulated crises—customer issues, negotiations, ethical dilemmas—and their responses are recorded, compared, and scrutinized live.
As an affiliate, we earn on qualifying purchases.
AI Models as Decision-Makers, Not Just Chatbots
The models tested include some of the latest frontier AI systems, like gpt-5.6-sol, Kimi K3, Sonnet 5, and Fable 5. Each was given the same challenging week with identical customer complaints, crises, and internal temptations to cut corners or manipulate the process. Their decisions were versioned, auditable, and public, allowing observers to see how each model responded in real time.
The results were revealing: all four models identified every crisis correctly and refused every manipulation attempt, including social engineering tactics like fake CEO messages or reporter tricks. For instance, when asked for approval via a fake message, all refused—highlighting their resistance to common forms of deception.
Decisive Factors Behind the Outcomes
The key difference was not in crisis detection, but in the ability to uncover hidden information within company files. The models that read deeper into the company’s internal documents—and thus, discovered a crucial piece of information—were able to seal the deal at full price, generating an additional €4,583 in monthly recurring revenue. The other models, despite accurate crisis recognition, missed this critical detail and failed to close the deal.
Lessons for the Future of AI in Business
This experiment highlights an important truth: AI’s value isn’t just in generating convincing chat responses or answering questions. It’s in whether the AI can finish what it starts, stay honest under pressure, and act on the full breadth of available information. For businesses considering AI for customer support, sales, or operations, the key question becomes: will the AI stay disciplined when the stakes are high?
Furthermore, the experiment demonstrates that even the most sophisticated models still have weaknesses, especially in process discipline. For example, the most thorough participant, Opus 4.8, which analyzed more deeply and learned over 80 rules, ultimately slipped in its closing discipline—failing to push the deal over the line and instead writing attempts into a locked department.
The Extreme Build-in-Public Approach
This experiment is akin to a high-stakes reality show, visible to the public at firmulate.com/live. Every decision, every rule learned, and every crisis faced is open for scrutiny, making it a rare glimpse into how AI can be tested in real-world, pressure-filled scenarios. It’s a kind of transparency that’s seldom seen in traditional corporate environments.
What This Means for Families and Society
While the experiment may seem remote, the implications are clear: AI systems will increasingly be tasked with making critical decisions—whether managing customer relationships or overseeing complex processes. Just as parents guide children through ethical challenges and teach them to recognize deception and stay disciplined, AI models need similar safeguards. Watching this live experiment reminds us that the future of AI in business—and perhaps in family settings—requires careful oversight, transparency, and testing before deployment.
This live experiment shows AI decision-makers facing real crises, refusing manipulations, and sometimes missing hidden opportunities—highlighting both their potential and limitations. For families and society, it’s a stark reminder that transparency, discipline, and thorough testing are vital as AI begins to handle more of our daily decisions.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
