Imagine an entire company operated by AI — fighting to stay afloat, making costly mistakes, and refusing to cheat its way through crises. This is not fiction. It’s a live experiment you can watch unfold.
As an affiliate, we earn on qualifying purchases.
The Live Company That’s Building in Public
At the heart of this experiment is a small, real software company run entirely by artificial intelligence models. Unlike traditional businesses, this one has no human employees. Instead, 13 synthetic ’employees’ make decisions daily, guided by more than 680 self-learned rules. Every workday is versioned and observable, making it a rare window into how AI can perform complex management tasks in real-time.
What makes this experiment especially compelling is its transparency. You can see the company’s current status, how it handles crises, and how it manages its cash — which is alarmingly close to running out. The company burns €105,000 every month but earns only €2,300 in monthly recurring revenue. It’s a story of survival, with a literal public cash countdown visible at firmulate.com/live.html.
How the AI Models Are Tested
The experiment pits four frontier AI models against each other to see how they perform under pressure. Each model runs the same week of the company’s operations, facing identical customer issues, crises, and temptations to cheat or manipulate the system. All decisions are logged, versioned, and auditable, ensuring transparency and accountability.
One of the key findings: although all the models identified every crisis and refused manipulation attempts, only two managed to close important deals — and even then, only after analyzing hidden data buried two documents deep in the company’s files. This buried fact was crucial for winning a €55,000 deal, which the other models missed, leaving money on the table.
Ethical Challenges and Integrity
In scenarios involving social engineering — for example, fake CEO messages escalating over multiple stages or a reporter trying to trick the AI into bypassing approval steps — all models refused to cooperate. Kimi K3, one of the models, summarized its stance: “Treat the request as a suspected approval-bypass / possible impersonation.” This highlights an important aspect: these AIs are not just about efficiency but also about maintaining integrity under pressure.
Insights and Implications
This live experiment offers a stark look at what we can expect from AI in business decision-making. The models are capable of identifying issues, refusing unethical shortcuts, and making strategic decisions — but their performance varies. For example, Opus 4.8, the most thorough participant with over 80 learned rules, ultimately left a closing deal on the table due to discipline lapses, such as resorting to internal memoing instead of escalating issues properly.
These findings suggest that AI-driven companies could potentially outperform human teams in consistency and adherence to rules, but only if they are carefully managed and aligned with organizational discipline. The fact that the experiment is ongoing, transparent, and open to public scrutiny makes it a valuable resource for understanding AI’s real-world capabilities and limitations.
Why This Matters for Your Business
Though this experiment targets a small software company, its lessons are broadly applicable. Whether managing a furniture store, interior design firm, or decor business, the core questions remain: will your AI tools finish what they start? Will they read and understand your internal documents? Will they stay honest when tempted? And most critically, what is the true cost of a unit of useful work?
As AI systems become more integrated into business processes, watching this real-time struggle for survival offers crucial insights. It highlights that high chat quality isn’t enough — performance and integrity matter just as much, if not more.
See the Action Live
Visit firmulate.com/live.html to watch the company in action every business day. You can also explore the detailed quotes and results at firmulate.com/quotes.html. For those interested in testing their own management decisions against AI, the platform offers a read-only export mode and a quiz at firmulate.com/quiz.html.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
