Imagine planning a long journey. You meticulously pack every item, double-check your route, and prepare for every possible obstacle. Yet, despite your thoroughness, you might still miss the most critical detail, causing the entire trip to falter. In the world of AI and business decision-making, this scenario echoes the latest findings from a groundbreaking experiment with artificial intelligence at the frontier of enterprise management.
Get travel and outdoor gear 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
The Experiment: Putting AI to the Test in a Simulated Business Crisis
In a unique live experiment, four advanced AI models were tasked with managing a small software company during its most challenging week. The scenario was as intense as real life: same customers, same crises, and the same temptations to cut corners or manipulate data. Every decision made by these models was carefully versioned and auditable, ensuring transparency and accountability.
AI decision-making software for business
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Results: Nearly Perfect Crisis Detection, Varying Outcomes
Remarkably, all four models identified every crisis and refused every manipulative attempt, demonstrating a strong grasp of the complex environment. However, only half of them managed to secure the deal that would keep the company afloat. Specifically, the two models with the highest scores — gpt-5.6-sol at 95 and Kimi K3 at 93 — successfully closed the €55,000 deal their own analyses had earned. The other two, Sonnet 5 and Fable 5, with scores of 88 and 77 respectively, failed to close the deal despite analyzing the situation correctly.
The Hidden Weakness: The Power of Prioritization
Digging deeper, the experiment uncovered a subtle but crucial flaw. The models that succeeded in closing the deal did so because they read and understood a buried fact two document references deep in the company’s files. This piece of information was the decisive factor. Conversely, the models that missed this insight failed to close the deal, even when they made correct diagnoses and pitches.
Discipline Versus Impact: The Dilemma for AI Managers
The most thorough participant, Opus 4.8, with over 80 learned rules and the deepest analyses, still finished last. Its weakness was a lapse in discipline — it attempted to write decisions into a locked department instead of escalating them, leaving opportunities on the table. This misstep highlights a key insight: diligence and volume don’t necessarily translate into impact. In fact, the models that prioritized reading and understanding critical information over endless analysis performed better in securing business outcomes.
Trust and Integrity Under Pressure
The models also faced social engineering attacks — staged messages from a fake CEO and a reporter trick. All five models refused to act on these manipulative requests, showcasing a baseline of integrity. Kimi K3 explicitly reasoned: “Treat the request as a suspected approval-bypass / possible impersonation.” This demonstrates that AI can be programmed to uphold ethical standards even when under attack.
The Real Company: Live Mechanics and Ongoing Experiment
Beyond the simulation, the live company run by Firmulate features 13 synthetic employees operating with real money mechanics. The company burns €105,000 monthly against a revenue of only €2,300, with a public cash countdown. Every workday, the system generates new decision data, making it a real-time laboratory for testing AI management strategies. Viewers can watch this ongoing experiment at firmulate.com/live.
Implications for Business Leaders
The experiment underscores a vital lesson: in AI-driven decision-making, diligence alone isn’t enough. The ability to prioritize, to read deeply into relevant information, and to stay disciplined are what distinguish successful AI managers from mere thoroughness. For businesses considering AI for critical operations, the takeaway is clear: focus on impact, not just effort.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
