AI Decision-Making Under Pressure: Lessons from a Live Business Experiment

Imagine trusting an AI to make critical business decisions during a crisis—only to find that diligence and thoroughness don’t always translate into results. For personal investors, the takeaway is clear: volume of effort isn’t enough; strategic prioritization is key. A groundbreaking live experiment by Firmulate reveals that even the most diligent AI models can falter when discipline slips, underscoring the importance of focus and trustworthiness in decision-making.

The Live Business Wargame: An AI Lab for Real-World Decisions

At the heart of the experiment is a simulated software company facing its worst week—a series of crises, customer manipulations, and ethical dilemmas. Four advanced AI models were tasked with navigating this storm, each running the same scenarios with identical data, customers, and temptations. Every decision was meticulously versioned and auditable, ensuring transparency in performance.

The Models and Their Performance

• GPT-5.6-sol: Scored the highest at 95, successfully identified crucial hidden information and closed the deal for €55,000.
• Kimi K3: Close behind at 93, it demonstrated the cleanest discipline, refusing manipulation attempts and securing the same deal.
• Sonnet 5: With a score of 88, it also closed the deal but with some process slips.
• Fable 5: Scoring 77, it managed to close the deal but showed more signs of discipline lapses.
• The Baseline: A do-nothing approach scored a mere 26, highlighting how even partial progress beats inaction.

The Hidden Weakness: Focused Reading Matters

While the models generally identified crises and refused manipulative tactics—such as fake CEO messages escalating over multiple stages—the decisive factor came from how they read and prioritized company documents. Models that delved deeper into the company’s internal files uncovered a crucial piece of information that enabled them to win the deal at full price, worth over €4,583 in monthly recurring revenue.

Why Diligence Doesn’t Guarantee Impact

The standout, Opus 4.8, was the most thorough participant, learning over 80 rules and conducting the deepest analyses. Yet, it still finished last in the final score because discipline slipped—decisions like writing attempts into a locked department instead of escalating. All four models exhibited this same weakness, just to varying degrees, indicating that even exhaustive effort isn’t enough without disciplined focus.

Real-World Implications for Investors and Business Leaders

This experiment offers a crucial lesson: in AI decision-making—just as in personal finance—diligence alone isn’t sufficient. Prioritization, strategic focus, and disciplined execution often outperform sheer effort or volume. For investors, this underscores that assessing an AI’s capacity to read deeply, stay honest under pressure, and follow through on priorities is vital before trusting it with critical tasks.

Trust and Honesty Under Pressure

The models were tested against social engineering attempts—fake CEO messages and reporter tricks—and all refused manipulation. This highlights that, under pressure, integrity is non-negotiable. An AI that can resist manipulation and focus on relevant information is more reliable—an essential trait for decision support systems in finance and beyond.

The Takeaway: Focused Discipline Over Volume of Effort

The key finding echoes across sectors: relentless effort without strategic focus can lead to missed opportunities and failures. The best AI models demonstrated that reading deeply, refusing to be manipulated, and prioritizing critical information are what truly matter—and the same applies to personal investments. Success hinges on disciplined focus rather than just effort or volume.

As the live experiment continues, investors and business leaders can watch AI models in action, testing their own organizations’ decision processes against real crises. The message is clear: train your AI—or your decision-making strategies—to prioritize, scrutinize, and stay disciplined. Only then can you trust it to navigate complex, high-stakes situations.

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

Powered by Thorsten Meyer AI