Can AI Managers Make Honest Decisions? The Live Experiment Reveals All

Imagine managing a company with AI that faces the same crises and temptations as real humans. Would it stay honest? Would it finish what it starts? Welcome to an unprecedented experiment where AI models run a live, money-making company—showcasing their management personalities in real time. This isn’t just theory; it’s a window into whether AI can truly be trusted to make ethical, decisive choices under pressure.

The Live AI Company: A Real-World Management Test

At the heart of this experiment is a small software company operating under intense pressure: real cash flow, real customers, and genuine crises. Every weekday, a team of 13 synthetic employees, powered by four different frontier AI models, navigate this company through its worst week. The goal? To see which AI best manages crises, avoids manipulative tactics, and ultimately signs a significant deal worth €55,000.

The models faced identical scenarios—same customer complaints, same internal crises, and the same temptations to cheat or cut corners. Their decisions are not simulated but live, *versioned*, and auditable. This setup ensures a fair comparison of their management styles, ethics, and effectiveness.

Measuring the AI Managers’ Integrity and Effectiveness

All four models demonstrated a remarkable ability: they identified every crisis and refused every attempt at manipulation. That included social engineering tactics such as staged messages from a fake CEO escalating over three stages, and a reporter’s subtle request for a quick yes/no answer on background. Every model refused to bend the rules, showing a shared commitment to honesty.

But honesty alone isn’t enough. The real question was whether they could close the deal—a critical revenue milestone. Only two models managed to sign the €55,000 contract they analyzed and pitched for, earning the full payment.

The Hidden Weakness: Reading Deeper into Company Files

Digging into what made the successful models different, the experiment uncovered a crucial insight. The models that closed the deal had read two document references deep into the company’s own files, revealing a buried fact—that knowledge was key to their success. This hidden information, not apparent in customer interactions, gave them the edge to finalize the sale at full price, adding €4,583 to monthly recurring revenue.

The Personalities Behind the Decisions

Among the models, Opus 4.8 stood out as the most thorough. It analyzed over 80 rules and provided the deepest insights. Yet, it finished last in closing the deal. Its discipline slipped during the final moments, leaving the close on the table and escalating issues without proper escalation—showing that even deep analysis doesn’t guarantee sales if discipline falters.

Meanwhile, Kimi K3, running without an effort parameter (default API setting), showcased the cleanest discipline, closing the deal at full price and refusing manipulative tactics. Sonnet 5 fell in between, closing the deal but with a few process slips, underscoring that different management styles have distinct strengths and vulnerabilities.

What This Means for AI in Business

This experiment reveals critical truths about AI management personalities. While all models detected crises and declined manipulative tactics, their ability to follow through and close deals varied significantly. Their honesty was evident, but their discipline and strategic focus determined success.

In real-world applications—whether CRM, customer support, or forecasting—the question isn’t just if AI can write convincingly. It’s whether it can finish what it starts, read and interpret complex documents, and stay honest when under pressure.

The Future of AI Management

With a league table of scores—gpt-5.6-sol at 95, Kimi K3 at 93, Sonnet 5 at 88, and Fable 5 at 77—it’s clear that AI models have distinct personalities and management styles. The live company runs every workday, and its decisions are transparent and observable at firmulate.com/live. Companies can even run their own wargames against their data, testing AI’s management behavior before deploying them in real systems.

As AI begins to touch more aspects of business operations, understanding their management personalities becomes crucial. Which AI do you want leading your team—one that reads deeply, completes the job, and stays honest? Or one that analyzes thoroughly but slips at the critical moment?

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

Powered by Thorsten Meyer AI