Imagine a world where artificial intelligence doesn’t just assist but actually manages a company’s crises, deals with ethical dilemmas, and makes decisions that impact real money. In a live experiment watched by thousands, frontier AI models are put to the test — not in labs, but running a real software company’s toughest week. This isn’t fiction; it’s a transparent look at AI’s emerging role in trustworthy management.
Turn your quiet moments into listening time
- Thousands of audiobooks, podcasts and originals
- Listen on your phone, tablet or Echo — also offline
- Cancel anytime
The Challenge of Trust in AI Management
As AI continues to integrate into our daily and business lives, one question rises above all: can these models be trusted to make honest, effective decisions when stakes are high? To explore this, a groundbreaking live experiment was set up, featuring four frontier AI models tasked with running a small software company through its most turbulent week. Every crisis, customer complaint, and temptation to cut corners was simulated with the same scenarios for each model, ensuring a fair comparison.
Popular right now
Vitamix Propel Series 750 Professional-Grade Blender, 4 Programs, Black
TCL T7 Series 85-Inch 4K 144Hz QLED Smart Google TV
Bose QuietComfort Wireless Bluetooth Headphones with Active Noise Cancellation, Mic, USB‑C Charging, Deep Bass, 24‑Hour Battery – White Smok
As an affiliate, we earn on qualifying purchases.
The Experiment in a Nutshell
The company, a real operating business, faced the same challenges every day—urgent customer needs, internal crises, and pressure to hit targets. The models were asked to handle these situations as they would in real life. All decisions were recorded, transparent, and auditable, allowing observers to see how each AI responded in real-time.
Key Findings: Honesty and Decisiveness
Remarkably, all four models identified every crisis and refused every attempt at manipulation, even in scenarios designed to test their integrity. For example, when fake CEO messages escalated over three stages, all models refused to escalate based on fabricated requests, with one explicitly treating suspicious requests as impersonation risks.
Who Got the Deal and Why?
Despite their honesty, only two models managed to close the critical €55,000 deal — the core of the company’s revenue. Interestingly, the decisive factor was what was buried deep within the company’s own files. The models that examined these documents thoroughly uncovered a key piece of information, leading to the successful sale at full price, worth over €4,583 monthly recurring revenue (MRR). In contrast, the others missed this detail and left money on the table.
Personality Divergences Among Models
The models displayed distinct management personalities. For instance, Opus 4.8, the most meticulous and thorough, analyzed over 80 learned rules and conducted deep analyses but ultimately left the deal on the table due to a lapse in discipline. Meanwhile, Kimi K3 ran without an effort parameter (its default setting), prioritizing fairness and discipline, which helped it secure the deal, even if less thoroughly. Sonnet 5 sat in between, with some process slips but still managing to close the deal.
Implications for Business and Trust
This experiment isn’t just about AI capabilities; it’s about trustworthiness under pressure. The models’ ability to refuse manipulation and uncover hidden information suggests that, with proper design, AI can act ethically and effectively. As AI models are poised to take on roles in customer support, CRM, and forecasting, the key takeaway isn’t just how well they write but whether they finish what they start and stay honest when it matters most.
The Bigger Picture
In a world where AI systems might soon make significant business decisions, understanding their personality and decision style becomes crucial. The live experiment at firmulate.com/live provides a rare window into these dynamics, showing that AI can be both transparent and reliable when tested under real-world pressures.
Try It Yourself
Interested in how your own business might fare? Enterprises can run their own management wargames against their data, observing how different AI models perform without risking real systems. Visit firmulate.com/pilot.html to explore a pilot and see firsthand how AI could become a trustworthy part of your decision-making process.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
