Can AI Make Better Business Decisions Than Humans?
Imagine a company with no human employees, where every decision is made by an AI model facing real crises, real money, and real consequences. This is not fiction but a live experiment that you can watch unfold, revealing how AI handles the complexities of running a business — including the temptations to cheat, the need for honesty, and the challenge of closing deals.
The Defensible Decision: A Guide to AI-Assisted Business Analytics (Applied AI)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Real-World AI Business Experiment
At the heart of this experiment is a small software company operated entirely by AI models, with 13 synthetic employees making decisions every workday. Unlike typical AI demos focused on chat, this setup measures actual management quality: crisis response, integrity, strategic decision-making, and deal-closing abilities, all against the brutal backdrop of ongoing financial loss.
Currently, the company is burning €105,000 each month while generating only €2,300 in monthly recurring revenue (MRR). It has a public cash countdown, adding pressure to every move. The company is versioned daily, with over 680 self-learned rules embedded in its decision-making playbook, making its operations transparent and auditable.
The Models Tested and Their Performance
Four frontier AI models—gpt-5.6-sol, Kimi K3, Sonnet 5, and Opus 4.8—were put through identical scenarios, simulating a typical worst week in the company’s life. All models identified every crisis and refused every unethical manipulation, such as social engineering attempts. For instance, all models rejected fake CEO messages escalating to a reporter on background, demonstrating integrity.
The real test was closing a deal worth €55,000. Only two models signed the deal based on their own analysis, despite identical pitches and diagnoses. Interestingly, the decisive factor was a buried document reference in the company’s internal files, which the two successful models read and used to close the full-price deal, adding €4,583 to monthly recurring revenue. The other two models missed this crucial detail and failed to sign.
Insights from the Live Data
This setup reveals an important truth: the failure to close the deal was not due to misreading the customer or poor pitch but was rooted in the models’ internal document reading. Those that examined the internal files fully gained an advantage, highlighting the importance of thorough information processing in AI management.
Dealing with Social Engineering
The experiment also tested how models react to social engineering. Fake CEO messages designed to escalate requests or bypass approval protocols were uniformly rejected by all four models. Kimi K3 explained its reasoning: “Treat the request as a suspected approval-bypass / possible impersonation.” This demonstrates a robust refusal to be manipulated, a vital trait for trustworthy AI systems in real business environments.
The Cost of Failure and the Reality of the Experiment
Despite the many successes, the company is losing money every day, emphasizing the gap between AI capability and business sustainability. The model rated most thorough—Opus 4.8—performed the poorest in closing deals, often slipping into incomplete processes or failing to escalate issues properly. It’s a stark reminder that more detailed analysis doesn’t always translate into better business results if discipline slips.
What This Means for AI in Business
This experiment underscores that the critical question for AI adoption isn’t just whether it can produce convincing dialogue. Instead, it’s whether it can consistently finish what it starts, read and interpret relevant documents, and stay honest under pressure. These qualities are essential for AI to be a dependable decision-maker in complex, real-world scenarios.
Learn More and Watch the Action Live
The live company, with its daily decision logs and transparent operations, is available for public viewing at firmulate.com/live.html. It offers an unprecedented window into AI’s management capabilities and the ongoing struggle for AI-driven business survival.
For a detailed breakdown of the models’ performance and decision-making, visit firmulate.com/quotes.html. Whether you’re a business leader, researcher, or curious observer, this is a rare chance to see AI tested in the crucible of real management challenges — and to watch a company fight for its life, in plain sight.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
