If you’ve ever wondered what it’s like to watch an entire company run by artificial intelligence—day after day, crisis after crisis—then the experiment at Firmulate might just be the most raw and revealing story you’ll see this year. It’s a real company, operating in real time, with a public cash countdown and 13 synthetic employees making decisions that matter. And it’s all happening in front of your eyes, as a live experiment in AI management.
Listen free for 30 days with Audible
Thousands of audiobooks and originals — cancel anytime.
As an affiliate, we earn on qualifying purchases.
The Live Company: A Real Business, Running in Public
At the core of this experiment is a small software company, watched daily as it battles to survive. Unlike typical corporate stories, this company has no human employees—only 13 synthetic decision-makers, each trained and guided by AI models. These models are not just chatbots; they’re running the entire operation. From managing crises to closing deals, every move is versioned, auditable, and on display for the world to see.
The company’s financials tell a stark story. It burns through €105,000 each month while generating just €2,300 in monthly recurring revenue. This relentless cash burn has a ticking clock—every day, the available funds dwindle, and the company’s survival hangs in the balance. Yet, despite the losses, this setup is an unprecedented window into how AI can be tasked with running a business in real time.
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.
The Experiment: Putting AI to the Test in the Worst Week
What makes this experiment unique is its rigorous testing environment. Four leading AI models, including GPT-5.6 and Kimi K3, faced identical challenges—same customers, same crises, same temptations to cut corners. Every decision was carefully tracked, with each version of the models’ responses recorded and made auditable.
The results are revealing. All four models identified every crisis and refused every manipulation attempt, including social engineering tactics like fake CEO messages and reporters’ tricks. Interestingly, the key advantage for some models was reading deeper into the company’s internal files. In fact, the decisive factor in winning a €55,000 deal was uncovering a buried document reference within the company files—something that other models missed. The models that read the full context secured the deal and increased the company’s monthly revenue by over €4,500.
The Reality of AI Decision-Making Under Pressure
Among the models tested, Opus 4.8 stood out for its thoroughness, analyzing over 80 learned rules and conducting deep analyses. Yet, it still finished last, leaving opportunities on the table and failing to escalate critical issues. This illustrates an important point: depth of analysis does not automatically translate into better outcomes, especially under real-time pressure and discipline lapses.
Another noteworthy finding is how models responded to social engineering. All five models, including Kimi K3, refused to be manipulated when presented with increasingly aggressive fake messages from a supposed CEO or reporter. Kimi K3’s reasoning was clear: treating such requests as potential impersonation or approval bypass attempts.
The Building Public in Its Most Extreme Form
This isn’t just a theoretical test. The company operates every business day, with decisions made in real time and every version stored for review. You can follow along at firmulate.com/live, where the ongoing story unfolds—showing daily decision logs, financial status, and even the model’s performance scores.
What makes this particularly compelling is that it demonstrates AI’s current strengths and weaknesses in a real-world context. It’s not a demo, nor a scripted scenario. It’s a real company losing money, fighting to stay afloat, with its management choices laid bare for all to see. Every decision made by the AI models is part of a larger experiment that explores the boundaries of automation, honesty, and business resilience.
Implications for the Future of AI in Business
This experiment reveals a key insight: the question isn’t just whether AI can generate convincing chat or perform well in isolated demos. The critical issue is whether AI can complete complex, goal-oriented tasks reliably—especially under pressure. Can it stay honest when faced with real crises? Can it uncover hidden facts that give it an edge? And, most importantly, can it finish what it starts, even when the financial stakes are high?
For companies contemplating AI-powered automation, the lessons are clear. Success depends on more than clever prompts or superficial interactions. It requires models that can read deeply, resist manipulation, and commit to completing critical tasks—like closing deals or managing crises—without cutting corners.
The Ongoing Story and How to Watch
For those interested in seeing this in action, the live site at firmulate.com/live offers a window into a company fighting for survival, powered by the most advanced AI models today. You can also explore the scores and analysis at firmulate.com/quotes, or participate in a quiz that challenges management decisions based on real AI responses.
In a world increasingly shaped by AI, this experiment provides a rare, transparent look at how these systems perform in a complex, high-stakes environment. It’s a story of innovation, struggle, and the relentless pursuit of testing what AI can truly do in business.
The live experiment at Firmulate offers a rare, transparent look into how AI models handle real business crises, decision-making, and integrity under pressure—highlighting both their potential and limitations as they battle to survive in real time.
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.
