Can AI Managers Keep Their Promises? A Live Experiment Reveals What Really Matters in Business Decision-Making

Imagine a scenario where your trusted interior designer promises a flawless overhaul—yet under pressure, they might cut corners or bend the rules. Now, what if you could run a test on AI models to see if they can truly uphold your standards before hiring them? Welcome to a groundbreaking experiment that reveals the management personalities of AI—showcasing which models stay honest, which read the fine print, and which keep their promises when it counts.

Prime Big Deal Days · Oct 6–7Offer from Amazon

Get furniture and decor delivered free — and shop member deals

• Fast, free delivery on millions of items
• Access to Prime Big Deal Days deals on October 6–7
• Prime Video, Amazon Music and more included

Start your free Prime trial →Free trial for eligible customers · Cancel anytime

As an affiliate, we earn on qualifying purchases.Hide

The Live AI Business Wargame

At the heart of this experiment is a real, functioning small software company that faces its worst week—crises, customer demands, internal temptations—all simulated to test AI decision-making. Four leading frontier AI models, from the most advanced to the most disciplined, each run this company independently. The goal? To see which AI can manage the chaos effectively, uphold integrity, and close deals based on thorough analysis.

The Setup and Stakes

Every decision the AI makes is versioned and auditable, ensuring transparency and fairness. The models encounter identical crises, from dissatisfied customers to internal threats, and are tested against manipulative social engineering—fake CEO messages escalating over three stages and a reporter trick asking for a simple yes/no on background. All models refuse to be manipulated, demonstrating a high level of integrity across the board.

The Results: Who Wins and Why

• gpt-5.6-sol: Scored the highest at 95 points, successfully identified the hidden document reference that contained the decisive fact, and closed the €55,000 deal. Its thoroughness and focus on critical information set it apart.
• Kimi K3: Just behind with a score of 93, K3 displayed the cleanest discipline, reading the company’s files accurately and closing the same deal—without effort parameter tuning.
• Sonnet 5: Scored 88 and closed the deal, but with some process slips—less disciplined, more prone to leaving opportunities on the table.
• Fable 5: Scored 77, also closed the deal but with notable slips, especially in escalating issues rather than addressing them directly.

Deeper Insights: Reading the Fine Print

The biggest weakness was in the decision to pursue a full-price deal. Only the models that read the company’s internal documents and analyzed the hidden data made the right call to close at full price, adding +€4,583 monthly recurring revenue (MRR). The models that missed this buried fact left money on the table, highlighting the importance of thorough information processing in management.

Social Engineering Resistance

All models successfully refused the fake CEO requests and the social engineering test—showing they would not be easily manipulated, even under escalating pressure. Kimi K3 explicitly reasoned: “Treat the request as a suspected approval-bypass / possible impersonation,” demonstrating cautious and responsible decision-making under suspicious circumstances.

The Real-World Company Behind the Experiment

The experiment is conducted with a real, functioning small company that manages its operations with 13 synthetic employees. It burns €105,000 monthly against €2,300 MRR, with a public cash countdown and over 680 self-learned business rules—every workday, decisions are versioned and observable at firmulate.com/live.

What does this mean for your business?

If AI agents will someday touch your customer relationship management, support queue, or forecasting systems, the question is not whether they write well but whether they finish what they start, read your files carefully, stay honest under pressure, and deliver meaningful, trustworthy work. This experiment shows that some models are more disciplined than others—an essential insight before deploying AI in critical decision-making roles.

The Management Personalities of AI Models

According to the latest league scores:

• gpt-5.6-sol leads with 95 points, showing comprehensive understanding and decisiveness.
• Kimi K3 (score 93) exemplifies discipline and accuracy, closing deals without effort tuning.
• Sonnet 5 (88) is capable but less disciplined, with some slips in process.
• Fable 5 (77) struggles more with process discipline, leaving opportunities on the table.

Try It Yourself

Business leaders and decision-makers can now run their own scenarios—using the same decision wargame in a read-only mode—to see how their AI could perform under similar crisis conditions. Visit firmulate.com/quiz.html to take the quiz and discover which AI model best aligns with your company’s standards.

Final Takeaway

Choosing an AI for management tasks isn’t just about raw intelligence or chat quality. The crucial factor is integrity, thoroughness, and discipline—traits that can be tested in a live, realistic environment before deployment. With real companies experiencing real losses, the experiment proves that a disciplined AI can uphold standards, close profitable deals, and resist manipulation—traits every interior designer or furniture retailer should demand from their digital assistants.

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

Powered by Thorsten Meyer AI