Can AI Management Models Win Over Real Business Crises? A Live Experiment Shows Surprising Results

In the high-stakes world of automotive repair shops and garages, trust and discipline are everything. What if the AI assistant you rely on for managing your team and customer relationships could actually be tested in the chaos — and prove its integrity? Welcome to a groundbreaking live experiment where AI models run a real, money-losing small business through its worst week, revealing not just what they do, but who they are.

The Experiment: Putting AI to the Test in Real Business Chaos

Imagine a busy garage facing the typical day from hell — angry customers, urgent repairs, internal crises, and temptations to cut corners. Now, think of four different AI management models each tasked with guiding this shop through the same tumultuous week. Every decision they make is recorded, transparent, and comparable. The goal? See which AI can handle crises, avoid manipulation, and close deals — just like a top-tier manager.

This is exactly what the live experiment at firmulate.com/live demonstrates. The models are tested against real business mechanics: managing 13 synthetic employees, balancing cash flow of €105,000 monthly against a meager €2,300 in monthly revenue, and navigating intricate rules learned over the course of many workdays.

The Results: Performance and Personalities Revealed

All four models successfully identified every crisis and refused every manipulation attempt — a crucial indicator of integrity. When it came to closing a €55,000 deal, only two models actually signed, despite their identical diagnoses and pitches. Interestingly, the decisive factor was a buried fact deep within the company’s files, not in the visible customer interactions. Models that accessed this information at the right moment secured the full deal, adding €4,583 in Monthly Recurring Revenue (MRR).

Among the competitors, GPT-5.6-sol scored the highest at 95, demonstrating full awareness and closing the deal flawlessly. Kimi K3, with a score of 93, also successfully closed the deal and exhibited the cleanest discipline — refusing to sign without proper analysis. Sonnet 5, scoring 88, did close the deal but showed some process slips. The least performing, Fable 5, scored 77, also closing, but with noticeable discipline lapses and leaving potential profit on the table.

Behind the Decisions: Personality Profiles of the AI Models

The experiment reveals that these models aren’t just code — they display distinct management personalities:

• GPT-5.6-sol: The most thorough and decisive, it prioritized reading the deep company files and closing full deals. It exhibits a meticulous, trustworthy persona.
• Kimi K3: The newcomer, with a focus on fairness, operated without an effort parameter, maintaining clean discipline, and making cautious decisions. It refused shortcuts and manipulation attempts confidently.
• Sonnet 5: Slightly more prone to process slips, but still able to close deals effectively. It seems more flexible but less disciplined under pressure.
• Fable 5: Despite closing deals, it showed discipline slips and left profits unclaimed, hinting at a more cautious but less disciplined personality.

Interestingly, all models refused to engage in social engineering tricks, such as staged CEO messages or reporter tricks, further demonstrating integrity under pressure.

Implications for Automotive and Garage Management

This experiment isn’t just about AI scoring; it’s about understanding whether AI can truly manage complex, real-world business scenarios — especially in sectors like automotive repair where trust, discipline, and integrity are vital. The fact that only some models closed full deals or read the deep files suggests that not all AI assistants are created equal in their managerial personalities.

For garage owners and automotive service managers, the takeaway is clear: choosing an AI isn’t just about its ability to generate chat or handle customer inquiries. It’s about how well it can uphold integrity, read and analyze critical information, and stay disciplined under pressure. These factors determine whether your AI can truly drive your business forward or leave profits and trust on the table.

Trying It Yourself: Run Your Own Wargame

Curious how your current management AI stacks up? You can run your own version of this scenario against a read-only export of your business at firmulate.com/pilot.html. It’s a safe, no-risk way to see if your AI of choice can handle real-world crises with honesty and discipline, before deploying it in your garage or repair shop.

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

Powered by Thorsten Meyer AI