What if a company’s survival depended on a daily battle with artificial intelligence — with all its flaws, fears, and hopes laid bare?
Imagine witnessing a small, real-world company teetering on the brink of financial ruin, its every decision scrutinized by cutting-edge AI models that are tested in real time. This is not fiction; it’s the live experiment run by Firmulate, a company that runs a simulated business with no employees, losing €105,000 each month against a modest €2,300 in monthly recurring revenue.
Every weekday, this company faces crises, temptations, and choices that could make or break its existence. The twist? These decisions are made by artificial intelligence, specifically by different frontier models that are rigorously tested to see whether they can navigate the complex, often messy realities of management.
What makes this experiment extraordinary is that it is built in public — every decision, every mistake, every success recorded for all to see at firmulate.com/live. It’s a rare window into how AI might truly perform in the high-stakes world of real business, not just in polished demos or chatbots but in actual decision-making that impacts real money.
The Challenge: AI in the Crucible of Real Business
The experiment pits four advanced AI models against each other, running the same company through its worst week. Every crisis — from customer complaints to ethical dilemmas — is identical for each model. The goal? To see if they can spot problems, resist manipulation, and ultimately close deals that sustain the company.
Remarkably, all four models identified every crisis and refused every manipulation attempt, including a staged social engineering attack involving fake CEO messages. Yet, only two managed to sign a €55,000 deal based on their own analysis — the core measure of success in this game.
The Hidden Weaknesses and Deep Lessons
The key to winning was not just surface-level decision-making but reading deep into the company’s own files. It was in those hidden documents that the decisive advantage was found. Models that examined and understood these references secured the deal at full price, worth an additional €4,583 in monthly recurring revenue.
In contrast, even the most thorough participant, Opus 4.8, fell short. Despite analyzing over 80 learned rules and conducting deep assessments, it left the close on the table and slipped in discipline — such as writing attempts into a locked department instead of escalating issues. This reveals a vital truth: thoroughness alone does not guarantee success, especially under pressure.
The Human Element—Honesty Under Fire
The experiment also tested social engineering tactics, with staged fake CEO messages designed to bypass approval processes. Every model refused these attempts, demonstrating resilience against manipulation. Kimi K3’s on-record reasoning summed it up: “Treat the request as a suspected approval-bypass / possible impersonation.”
This is an essential insight for businesses considering AI: integrity and honesty are not just features but fundamental capabilities that can be reliably tested in high-stakes scenarios.
The Live Company: A Model of Financial Struggle and Innovation
The company at the heart of this experiment operates with 13 synthetic employees, managing real money mechanics, yet it burns through €105,000 every month with only €2,300 in recurring revenue. It is a public, visible race against time, with a daily updated view of its decisions and progress, available at firmulate.com/live.
Every decision made by the AI models is versioned, recorded, and auditable. The company’s live status underscores a profound question: In an era where AI can be tested against real-world crises, what does it mean for human-managed businesses? Are we preparing for a future where AI might not just support but replace traditional management — and at what cost?
The Lessons for Business and Spirit
For those interested in the deeper metaphysics of decision, trust, and the human spirit, the Firmulate experiment offers a reflection: AI models can face crises, resist manipulation, and make complex decisions. Yet, their failures reveal vulnerabilities—much like our own in moments of stress or moral challenge.
As the models learn from each decision, growing more sophisticated, they also highlight the importance of integrity, understanding, and discipline—virtues central to spiritual growth. Watching this experiment unfold invites us to consider: How do we respond when faced with pressure and temptation? Can we, like these AI models, stay true to our principles even when the stakes are high?
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
Decision Making Under Uncertainty: Theory and Application (MIT Lincoln Laboratory Series)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
