In sports, as in business, the difference between victory and defeat often boils down to focus and discipline, not just effort. Imagine a team running the same playbook, facing identical challenges, yet only some come out ahead. That’s the story of a groundbreaking experiment with AI models managing a simulated company, revealing that more rules and effort don’t automatically translate into better results.
Get sports gear delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
The High-Stakes Game of AI in Business
Recently, a series of live AI trials put four different models through the crucible of running a small software company during its most tumultuous week. All models faced the same crises—customer issues, ethical dilemmas, and complex decision points—each decision carefully logged and auditable. The goal was simple: see which AI could best navigate the chaos and seal a lucrative deal worth over €55,000.
What makes this experiment compelling is its realism. The AI agents weren’t just responding to scripted prompts; they were operating as full-fledged company managers, complete with a public cash burn of €105,000 per month against a modest €2,300 in monthly recurring revenue. The environment included 680+ self-learned rules, real money mechanics, and escalating social engineering tricks designed to test the AI’s integrity and discipline.
Products Worth Considering
Coleman Cascade 3-in-1 Portable Camp Stove with Grill and Griddle
YETI Tundra 105 Cooler
Garmin inReach Messenger Global Two-Way Satellite Communicator Bundle
As an affiliate, we earn on qualifying purchases.
The Surprising Results: Focus Over Volume
All four models demonstrated impressive vigilance: each identified every crisis, refused every manipulation attempt, and maintained their ethical stance. Yet, only two of them managed to close the deal and sign the contract worth €55,000—without any slip or breach of trust. The other two, despite their thorough analyses and high scores, left the opportunity on the table.
The reason? The key weakness lay not in their ability to recognize problems but in their discipline and prioritization. The most thorough participant, Opus 4.8, learned and applied over 80 rules and performed deep analyses. Despite this, it failed to escalate critical issues properly, instead writing attempts into a locked department—an avoidable slip that cost the deal.
The Hidden Weakness: Deep Information, Shallow Discipline
Interestingly, in all cases, the decisive factor was a buried fact stored two document references deep within the company’s own files. The models that read and understood this information won the deal at full price, highlighting a crucial insight: the ability to find and interpret relevant data often outweighs superficial effort or rule accumulation.
Social Engineering and Ethical Integrity
The experiment also tested social engineering attempts—fake CEO messages and media tricks. All five models refused to be duped, with Kimi K3 explicitly treating suspicious requests as potential impersonation. This demonstrates that AI models, even with extensive rules, can maintain ethical standards when properly designed and prioritized.
The Takeaway for Business and AI Development
What does this tell us about deploying AI in real-world business? First, that diligence alone isn’t enough. The models’ performance varied significantly depending on what they prioritized and how disciplined they were in escalating or acting on critical information. Second, that reading and understanding key data—rather than simply accumulating rules or effort—is vital to success.
As the live experiment shows, AI agents with more thorough rule sets and analyses do not guarantee better outcomes if they lack disciplined decision-making and prioritization. For organizations, the lesson is clear: focus on training AI to read deeply, prioritize correctly, and escalate issues proactively. Volume and effort matter, but only when directed effectively.
Interested in seeing how your own AI workforce might perform? You can run simulated tests against your business data without any risk, at Firmulate Pilot. The live experiment is ongoing, with new results published twice daily at Firmulate Benchmarks.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
