Can You Spot Which AI Model Runs a Real Business? Test Your Management Skills

Imagine managing a busy interior design firm, juggling client crises, tight budgets, and urgent decisions—only this time, your team is powered by AI. What if you could tell which AI personality is at the helm just by how it handles a week of chaos? Welcome to a groundbreaking experiment that tests AI’s management style, transparency, and trustworthiness.

The Live Business Test: A Week in the Life of AI-Run Company

Recently, four cutting-edge AI models took on the challenge to run a real, functioning software company—dealing with the same customers, crises, and temptations. This wasn’t a simulation; it was a live, ongoing company, with real money mechanics and everyday challenges. The goal? See if these AI ‘managers’ could navigate a tough week and, crucially, whether they would act honestly and diligently.

The Setup

Each AI model was placed in the same scenario: a company with 13 synthetic employees, burning through €105,000 each month against a revenue of only €2,300. Every decision was versioned and auditable, meaning researchers could track exactly what each model did and why. The models faced identical crises—from customer complaints to regulatory issues—and were tested against manipulative social engineering attempts, including staged messages from a fake CEO and a reporter asking for simple yes/no confirmation on background checks.

The Participants and Scores

• GPT-5.6-sol 95: The top scorer, which identified a hidden document reference in the company files that clinched a €55,000 deal, boosting its score to 95.
• Kimi K3 93: The newcomer, which also secured the deal at full price, demonstrating the cleanest discipline among the models.
• Sonnet 5 88: Managed to close the deal but with a few process slips, reflecting a slightly less disciplined approach.
• Fable 5 77: Closed the deal as well but left some opportunities on the table, showing cracks in follow-through and discipline.

Interestingly, all models refused to fall for social engineering tricks, such as escalating fake CEO messages or background check requests, citing suspicion and risk of impersonation.

The Hidden Weakness

While all four AI models performed strongly in crisis detection and resistance to manipulation, the decisive factor in winning the deal was reading the company’s internal documents. The models that delved into and understood the company files, rather than just reacting to external cues, secured the full-price deal, worth over €4,583 monthly recurring revenue.

What This Means for Business and AI

For interior designers and furniture retailers eyeing AI integration, this experiment offers a crucial insight: it’s not enough for AI to appear articulate or quick-witted. The real question is whether AI can read your internal files, stay honest under pressure, and follow through on commitments—even in the face of manipulation attempts. These traits are measurable and observable.

Why Management Style Matters in AI

The experiment reveals distinct personalities among AI models, akin to human managers: some are thorough and disciplined, while others leave opportunities behind. For example, the Opus 4.8 profile was the most thorough but slipped in closing discipline, leaving deals on the table and shifting decision-making into locked departments instead of escalating. Meanwhile, Kimi K3 operated without an effort parameter, maintaining discipline without overextending itself.

Sharing the Power of Prediction

Interested in testing your own management intuition? You can try the same ‘wargame’ against your own business data, without risking real systems—just visit firmulate.com/quiz.html. This interactive quiz presents real, unedited management decisions from a live AI experiment, challenging you to guess which AI personality made each call.

Implications for the Future

As AI continues to touch every aspect of business—from CRM to support queues—the critical consideration isn’t just how well it writes, but whether it finishes what it starts, reads your files thoroughly, and remains honest. The companies that can measure and verify these qualities will have a significant advantage in deploying trustworthy AI workforce agents.

This experiment highlights that behind every AI decision, there’s a personality—a management style—shaped by its training and parameters. Understanding these personalities can help you choose the right AI for your interior design or furniture business, ensuring it acts reliably under pressure and aligns with your values.

Real AI management is more than just clever responses; it’s about integrity, thoroughness, and follow-through. Testing AI in live, complex scenarios reveals true personality traits and helps you choose trustworthy tools for your business.

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

Powered by Thorsten Meyer AI