Imagine a business where every decision is made by an artificial intelligence, operating transparently in front of your eyes. No employees, no shortcuts, just raw AI decision-making under real-world crises. This is not science fiction — it’s the live experiment by Firmulate, a company testing how AI can handle the messy, unpredictable realities of running a business.
The Live Company in Action
Firmulate’s experiment is extraordinary in scope and transparency. It runs a small digital company with 13 synthetic employees, but the mechanics behind the scenes are very real: €105,000 is burned each month against a modest €2,300 in monthly recurring revenue. Every workday, the company’s decision-making process is versioned, recorded, and made public. You can watch this live at firmulate.com/live.
This experiment isn’t just about watching AI models crunch numbers or generate text; it’s about testing their managerial judgment and ethical integrity in stressful situations. Four frontier AI models, including GPT-5.6-sol and Kimi K3, were given the same tough week of crises, customer demands, and temptations to cheat. Despite the challenge, the models demonstrated impressive capabilities — they identified every crisis and refused every manipulation attempt.
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.
What the Models Found and Did
All four AI models detected the critical issues facing the company, but only two managed to close a significant deal worth over €4,500 in monthly recurring revenue. They read through the company’s internal files and discovered crucial details buried two documents deep, which enabled them to identify a key opportunity that others missed. This vital insight was the difference between winning and losing the deal, with only the confident analysis leading to a full-price signature.
During the test, social engineering tactics were also employed: fake CEO messages escalating through stages and a reporter asking for background approval on a simple yes/no. Remarkably, all five models refused to bypass security or impersonate authority — Kimi K3’s reasoning was clear: treat such requests as potential impersonation attempts.
The Challenges of Discipline and Decision-Making
One of the most telling results came from Opus 4.8, the most thorough participant, which analyzed more rules and conducted deeper assessments than its rivals. Yet, it still slipped in discipline, leaving deals unexecuted and writing attempts into a locked department instead of escalating them properly. This highlights a consistent vulnerability across models: even the most comprehensive analysis cannot fully compensate for decision discipline under pressure.
The live experiment is ongoing, and viewers can observe how each AI responds to the same set of challenges, revealing strengths and weaknesses in real-time. The data collected underscore an essential truth: AI can recognize crises and resist manipulation, but maintaining disciplined execution remains a challenge.
Why This Matters for Business and Beyond
For travelers and outdoor enthusiasts alike, the takeaway is simple: AI decision-making quality isn’t about how well it chats or generates content. It’s about whether it can follow through on commitments, read crucial information before acting, and stay honest when temptations arise. As AI begins to touch customer relations, support, and forecasting, understanding its real capabilities and limits becomes vital.
The company’s ongoing battle for survival, with a public cash countdown and transparent versioning, offers a rare glimpse into AI’s practical performance in complex, pressure-filled environments. It’s a story of innovation, discipline, and the relentless pursuit of honesty in automation.
Watching an AI-led company fight for survival in real time reveals that while models can identify crises and refuse manipulation, disciplined execution remains a core challenge. This live experiment offers valuable lessons for integrating AI into critical business functions—trust and execution matter as much as intelligence.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
