Diligence Alone Isn’t Enough: What AI’s Last-Place Finish Reveals About Impact and Prioritization

Imagine hiring an AI assistant that meticulously follows every rule, deflects every scam, and even reads your hidden files. Yet, it still fails to close the deal that matters most. For relationship or business success, diligence without prioritization can be a costly mistake—something AI models are now revealing in real time.

How AI Testing Unveils the Limits of Diligence

The latest experiment from Firmulate puts four of the most advanced AI models through their paces—simulating a small software company’s worst week, complete with crises, temptations, and high-stakes decisions. The goal? To see if these AI systems can navigate chaos, uphold integrity, and deliver tangible results—like sealing a €55,000 deal.

Equal Conditions, Different Outcomes

All four models—ranging from the top-scoring gpt-5.6-sol to the relatively weaker Sonnet 5—faced identical scenarios. Each identified every crisis and refused every manipulation attempt, demonstrating a shared ability to recognize threats and maintain honesty. Yet, only two managed to close the deal based on their own analysis, while the other two fell short, leaving the most crucial opportunity on the table.

Surprising Weaknesses Hidden in the Files

The key difference? The successful models found critical information buried two document references deep within the company’s files—details that, if read, revealed the opportunity needed to seal the deal at full value. The ones that skipped over that layer missed out entirely, leaving €4,583 in monthly recurring revenue (MRR) unrealized.

Refusing to Be Manipulated

Social engineering tactics—like staged CEO messages and reporter tricks—were uniformly rejected by all models. Kimi K3, the standout in fairness, even flagged suspicious requests as possible impersonation, demonstrating an understanding of manipulation risk that goes beyond surface-level checks.

Implications for AI in Business and Relationships

This experiment underscores a vital truth: diligence and comprehensive analysis are not enough. An AI’s ability to impact depends heavily on its capacity to prioritize the most valuable information and actions. A model that is thorough but unfocused risks missing the big opportunities—an issue equally relevant to relationships and business.

The Opus 4.8 Profile: Deep, Yet Disconnected

The most thorough participant, Opus 4.8, learned over 80 rules and conducted deep analyses. Nonetheless, it finished last because discipline slipped during the final moments, leaving the close on the table and the decision process scattered. Its failure was not due to lack of diligence but a lapse in prioritization—highlighting that volume without focus can be counterproductive.

The Broader Lesson: Prioritize Impact, Not Just Effort

All four models displayed this pattern. Even with extensive rules and deep analysis, the weakest failed to act decisively when it mattered most. For businesses and individuals alike, this suggests that success hinges less on sheer effort and more on strategic focus. It’s not enough to do everything; you must do the right things, at the right time.

What This Means for Human and AI Decision-Making

In real life, whether choosing a partner or a business partner, the same rules apply. Recognizing key signals buried in complex information, resisting manipulation, and acting decisively are critical. Diligence, while vital, must be coupled with prioritization to truly make an impact.

Watch the Live Experiment

Curious how these AI models perform in real-time? The live experiment from Firmulate offers a window into how AI handles complex, high-stakes scenarios—each decision versioned and auditable. It’s a rare opportunity to see AI not just mimic human conversation but emulate entire company operations under pressure.

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

Powered by Thorsten Meyer AI