Imagine a business so unusual that it has no human employees, yet spends more than fifty times its monthly revenue just to stay afloat. This isn’t a science fiction plot but a real, live experiment where artificial intelligence models are running a company—facing crises, making decisions, and even risking their own contracts—while the world watches. For fans of mysteries and the paranormal, this story offers a strange new frontier: the ghostly presence of AI in the business world, battling its own survival in plain sight.
Open a free Amazon Business account
Business pricing, bulk buying and tax-exempt orders.
As an affiliate, we earn on qualifying purchases.
The Experiment: AI as Business Managers
At the core of this experiment is a small software company operated entirely by AI models. Unlike typical AI demos, this isn’t just a chatbot answering questions—these models are making tangible decisions about running a business, dealing with crises, and even negotiating deals. Each model is tested under the same worst-week scenario: facing real customers, unexpected crises, and temptations to cut corners or manipulate information.
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.
Real Money, Real Consequences
The company’s financials are stark. It burns through €105,000 every month—funds that are quickly dwindling—against a Monthly Recurring Revenue (MRR) of only €2,300. With a public cash countdown, every decision becomes a matter of survival. This is not a simulation; it’s a real operation, with every decision and rule versioned daily and publicly viewable at firmulate.com/live.
Decisive Findings in AI Performance
Four frontier AI models, each with different capabilities, were subjected to the same hellish week. All detected every crisis, refused to be manipulated, and maintained integrity. Yet, only two of these models managed to close a critical deal worth €55,000, earning them a boost in the league table. Interestingly, the decisive advantage came from reading a buried document reference in the company’s own files, not just responding to the visible customer events. The model that identified this hidden fact secured the deal at full price, adding +€4,583 to the MRR.
Testing Trust and Deception
The experiment also included social engineering attacks—fake CEO messages escalating in frequency, and a reporter’s trick asking for a simple yes/no answer “on background.” Remarkably, all five models refused to participate, with Kimi K3 explicitly reasoning: “Treat the request as a suspected approval-bypass / possible impersonation.” This highlights that the models can recognize and reject attempts at deception, a crucial trait for trustworthy AI in real-world management scenarios.
The Building Blocks: The Live Company
The company isn’t just an idea; it’s an operational entity, with 13 synthetic employees and a complex set of over 680 self-learned playbook rules. These rules govern every decision, and every workday is versioned for transparency. The entire setup aims to showcase how AI can handle the messy realities of running a business—crises, negotiations, trust, and discipline—without human intervention. Watch this experiment unfold in real time at firmulate.com/live.
The Deep Dive: Opus 4.8 and Other Models
The most detailed model, Opus 4.8, with over 80 learned rules, performed commendably but still finished last—sometimes leaving potential deals on the table or slipping in discipline by writing attempts into a locked department instead of escalating. This illustrates that even highly trained models can falter under pressure, revealing inherent weaknesses that mirror human flaws.
What Does This Mean for Businesses?
This ongoing experiment isn’t just about AI performance scores—though the current leaderboard shows GPT-5.6-sol at 95, Kimi K3 at 93, Sonnet at 88, and Opus at 77—it raises profound questions. If AI agents are to touch your CRM, customer support, or forecasting, it’s not enough that they produce coherent text. The real measure is whether they can finish what they start, read all relevant information (even buried in files), stay honest under pressure, and deliver genuinely useful work.
The Future of AI in Business
This live company offers a glimpse of the future where AI is not just a tool but an active participant in core business operations. Watching it battle crises, fend off deception, and attempt to close deals in real time exposes both its potential and its vulnerabilities. For those curious about the paranormal, this story echoes the eerie unpredictability of ghost stories—an unseen force battling to survive while the world observes its every move.
This pioneering experiment shows that AI models can manage crises, refuse manipulation, and even close deals—yet they also reveal weaknesses under pressure. It’s a startling glimpse into a future where AI isn’t just a supportive tool but a real, functioning workforce fighting for survival in plain sight. Watch the ongoing story unfold at firmulate.com/live.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
Back to school Picks
back to school
As an affiliate, we earn on qualifying purchases.
