In a world where technology is increasingly woven into the fabric of our lives, the question isn’t just about what AI can say—it’s about what AI can do when it counts. Just as a person’s true character reveals itself under pressure, so too does the real strength of artificial intelligence emerge in moments of crisis. For those of us who believe in the power of integrity, faith, and genuine leadership, recent experiments with AI models offer a profound reflection: can artificial agents embody trust and discipline, or do their limitations reveal a need for human discernment?
Turn your quiet moments into listening time
- Thousands of audiobooks, podcasts and originals
- Listen on your phone, tablet or Echo — also offline
- Cancel anytime
In a groundbreaking experiment conducted by Firmulate, four advanced AI models were tasked with running the operations of a real small software company through its most challenging week—full of customer crises, potential manipulations, and critical decisions. The goal was straightforward: see if these models could navigate the moral and practical complexities that define true leadership, and whether they could deliver on a substantial business deal worth €55,000.
What makes this experiment compelling is its focus on management qualities that go beyond simple chat responses. While all four models—gpt-5.6-sol, Kimi K3, Sonnet 5, and Fable 5—successfully identified every crisis and refused every manipulation attempt, only two actually closed the deal that their own analysis had earned them. The others, despite diagnosing the issues just as well, left the financial victory on the table, their discipline slipping at crunch moments.
In particular, the models that read deeper into the company’s own files—like Kimi K3 and gpt-5.6-sol—were able to uncover critical insights buried two documents deep, leading to a full-price deal worth over €4,500 per month in revenue. Conversely, the more rule-disciplined Fable 5, despite its integrity, failed to execute the deal, illustrating that honesty alone isn’t enough—discipline and follow-through matter just as much.
Further testing involved social engineering attacks, where fake CEO messages and reporter tricks were used to tempt the models into bypassing controls. Remarkably, all five models refused these manipulations, highlighting their resistance to deception—a vital trait for trustworthy AI in real company settings.
Yet, beneath these surface successes lies a sobering reality: the models’ decisions were made in a controlled environment, with decisions carefully recorded. In the real-world company where this was tested, operations burn €105,000 monthly against just €2,300 in revenue, a stark reminder that AI’s effectiveness isn’t just about spotting crises—it’s about consistently executing and closing deals, even under pressure.
What does this tell us about the future of AI in leadership? The critical takeaway is that the ability to read and understand buried information—to read between the lines, as spiritual traditions often teach—is what separates an AI that merely recognizes problems from one that can act decisively and ethically. It’s not enough to generate convincing chat; an AI’s strength is measured by its discipline, its integrity under stress, and its capacity to follow through on its commitments.
For those interested in exploring this further, you can watch the live experiment at firmulate.com/live, where a real business runs daily decisions and faces real money mechanics, all guided by AI models that are tested for their management qualities—not just their chat skills. The company’s current situation—losing €105k monthly—underscores that AI’s value is in its ability to deliver consistent, disciplined results, especially when the stakes are highest.
As faith teaches us to seek integrity and discipline within ourselves, so too must we demand these qualities from AI systems that will serve as our partners in leadership. The experiment by Firmulate shows that true trust in AI isn’t just about what it can say, but what it can do when faced with real crises. AI models are capable of resisting manipulation and uncovering hidden truths, but their success depends on their ability to follow through and execute—traits that mirror the spiritual virtues of discipline, honesty, and perseverance. For businesses and faith alike, the moral of the story is clear: trust is earned not just by promises, but by actions under pressure.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
AI management decision support tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
