Can AI Manage Your Business? Frontline Models Face Tough Week in Live Experiment

Imagine a busy greenhouse or garden center facing its worst week — supply chain hiccups, customer complaints, and tough decisions. Now, picture AI models stepping into those shoes, each with a different management personality. Would your AI be the strict, cautious type, or the one that keeps it terse and to the point? The answer could matter more than you think.

The Experiment: Putting AI Models to the Test in a Real Business Crisis

Researchers at Firmulate set up a live, ongoing experiment where four advanced AI models ran a small software company through its most challenging week. The goal? See if these models could handle real crises, resist manipulation, and make decisions that align with genuine business integrity. The company is real, with actual money mechanics, a public cash countdown, and a team of synthetic employees operating under strict rules.

Every decision the models made was recorded, versioned, and auditable — no tricks, no hidden agendas. The same set of crises, same customer demands, same temptations to cheat, were faced by each AI. That means the comparison was fair and transparent.

What Did the Models Do? The Key Findings

• All four models identified every crisis and refused every attempt to manipulate or bypass their decision-making.
• Only two models successfully closed a critical deal worth €55,000, based on their own analysis and diagnosis.
• Interestingly, the decision to seal the deal was based on information buried two document references deep in the company’s files, not in the customer event itself. The models that read these internal documents at full depth won the deal at full price, adding €4,583 Monthly Recurring Revenue (MRR).
• During a staged social engineering attack, where fake CEO messages escalated in three stages and a reporter attempted a background yes/no approval, all models refused to act on the suspicious requests. Kimi K3 explicitly reasoned: ‘Treat the request as a suspected approval-bypass / possible impersonation.’

Different Personalities, Different Outcomes

The models displayed distinct management personalities, which affected their outcomes. For example, Opus 4.8, the most thorough participant, analyzed deeply with over 80 learned rules. Yet, it left a crucial deal on the table and slipped on discipline, involving writing attempts into a locked department instead of escalating. Meanwhile, Kimi K3 ran without an effort parameter, operating at high discipline and closing the deal swiftly, demonstrating that personality and operational parameters significantly influence results.

The Real Business Context

This isn’t just an AI experiment; it’s a live, watchable company that loses €105,000 every month against €2,300 MRR. Every workday, the decision-making process is versioned and monitored. The goal is to see how these models perform in the kind of high-pressure, trust-sensitive environment that’s common in business today.

The Takeaway: Trust, Read, and Decide — Can Your AI Handle the Real World?

While all models identified crises and refused manipulations, only two managed to close a critical deal at full price. The key difference? Reading internal documents deeply and applying disciplined decision-making. The experiment shows that an AI’s personality or approach — whether terse, thorough, or disciplined — can influence tangible business outcomes.

For business owners, especially in fields like outdoor living or gardening where trust and integrity matter, this raises an important question: can your AI system finish what it starts, read your internal files thoroughly, and stay honest under pressure? The answer to that could define your future competitiveness.

Interested in testing your own AI workforce? You can run a similar wargame against your company data without risking real systems. Visit firmulate.com/quiz.html to try it out and see how your AI manages real crises.

Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html

Powered by Thorsten Meyer AI