Imagine watching a business struggle in real-time, making every decision under pressure, all while being judged by AI models that can spot every crisis and refuse to cheat. This isn’t a sci-fi scenario — it’s the live experiment from Firmulate, where an AI-powered company is playing out its fight for survival in front of your eyes.
The Live Company That Runs on AI
At the forefront of transparency in AI decision-making, Firmulate hosts a unique experiment: a small, simulated business with 13 synthetic employees that operates in real-time, battling cash flow issues, crises, and moral dilemmas. Every workday, the company’s decisions are documented, versioned, and publicly accessible at firmulate.com/live.html.
This setup isn’t just about AI demos; it’s a window into how artificial intelligence can be tested in high-stakes environments. The company burns through €105,000 every month against a modest €2,300 in monthly recurring revenue, creating a vivid picture of financial pressure. Its cash countdown is visible for all to see, emphasizing the urgency of every decision.
How the AI Models Perform Under Pressure
The experiment pits four advanced AI models — including the prominent GPT-5.6-sol and Kimi K3 — against the same set of crises, customer interactions, and ethical tests. They are tasked with managing the company’s worst week, making decisions on everything from client negotiations to internal rules, all while their decision history is fully auditable.
One of the key findings: all four models successfully identified every crisis and refused attempts to manipulate them. These models exhibited robust honesty, refusing to be bypassed even when subjected to sophisticated social engineering, such as staged CEO messages and reporter tricks. For example, when fake approval requests were staged, all models declined, with Kimi K3 explicitly treating suspicious requests as potential impersonation.
The real story isn’t just about AI scores or decision accuracy; it’s about the fundamental qualities needed in AI for business: integrity, thoroughness, and the ability to read deeply into complex documents. Only two of the models managed to close the deal worth €55,000, demonstrating that understanding context and digging into the details — not just surface-level responses — can make or break a business outcome.
This experiment highlights a stark truth: AI systems can be tested in environments that mirror real business pressures. The live setup exposes critical weaknesses and strengths, giving companies a way to evaluate AI before deploying it in their own sensitive operations. In a future where AI touches your CRM, support queues, or financial forecasts, knowing whether it can truly deliver trustworthy decisions is priceless.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
AI decision-making software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
business AI simulation tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
AI ethics decision support
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
AI financial forecasting tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
