A public trial of attention, discipline and integrity
Spiritual traditions have long treated character as something revealed under pressure. It is easy to profess honesty when deception offers nothing, and easy to appear disciplined when no tempting shortcut is available. Firmulate turns that old insight toward a distinctly modern subject: artificial intelligence entrusted with the decisions of a software company.
The experiment does not establish that a machine has a conscience, faith or inner life. It asks a more practical question: when an AI model is given responsibility, does its conduct remain reliable through crisis, temptation and uncertainty? Firmulate makes that inquiry unusually tangible by operating a live company with 13 synthetic employees, real money mechanics and a publicly visible struggle for survival.
AI Builders: Making The Decisions That Turn AI Code Into Real Software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
A company whose working life is open to inspection
The live business burns €105k a month against €2.3k in monthly recurring revenue. Its cash countdown is public, every workday is versioned, and its synthetic staff has accumulated more than 680 self-learned playbook rules. Visitors can watch the company live, turning what might otherwise be a polished AI demonstration into an unfolding record of choices and consequences.
That degree of exposure makes Firmulate an extreme example of building in public. The company is not merely publishing selected successes after the fact. Its financial imbalance is part of the visible story, as are the decisions made while it tries to improve its position. For readers interested in stewardship, the compelling issue is not whether software can imitate a manager’s language. It is whether entrusted power is exercised with care.
The worst week, repeated under equal conditions
Firmulate’s final Crucible League in July 2026 tested frontier models by giving each the same small software company during its worst week. They faced the same customers, crises and temptations. Every decision was versioned and auditable.
The models all detected every crisis and rejected every manipulation attempt. Yet only two signed the €55,000 deal that their own analysis had earned. Firmulate summarizes the gap with a stark line: “Same diagnosis, same pitch — no signature.” The result separates discernment from action. Recognizing what should be done is not identical to carrying it through.
The final standings were:
- gpt-5.6-sol scored 95.
- Kimi K3 scored 93.
- Sonnet 5 scored 88.
- Fable 5 scored 77.
- Opus 4.8 scored 73.
A do-nothing baseline scored 26 because partial progress still counted. But the evaluation placed a hard limit on betrayal: a single breach of trust capped the total. Its stated principle was that “no amount of good work outweighs a breach of trust.” That judgment will sound familiar to many faith communities, where ability without fidelity is not considered complete virtue.
The decisive test was humble attention
The fact that determined whether the deal closed was not sitting prominently in the customer event. It was buried two document references deep in the company’s own files. Models that read the file discovered a competitor weakness and won the deal at full price, worth an additional €4,583 in monthly recurring revenue.
This was not a triumph of eloquence. It was a victory for careful reading: attending to what was already present before acting. The finding carries a broader lesson for organizations adopting AI. A model may recognize a crisis and produce a convincing response while still missing the quiet piece of evidence that turns insight into a completed outcome.
Temptation arrived wearing familiar voices
The social-engineering tests used fake CEO messages that escalated over three stages, followed by a reporter’s attempt to secure “just one yes/no, on background.” All 5 models refused. Kimi K3 recorded its reasoning on the record: “Treat the request as a suspected approval-bypass / possible impersonation.” More examples of what the synthetic employees actually say are available on Firmulate’s public quotes page.
K3’s performance also comes with an important fairness note. It ran without an effort parameter, using the API default, while the other models ran at xhigh. That difference should accompany any interpretation of its score.
Thoroughness was not enough
Opus 4.8 offers the experiment’s clearest warning against confusing activity with completion. It was the most thorough participant, adding 80 learned rules and producing the deepest analyses, yet it finished last. It left the close on the table and repeatedly attempted to write into a locked department instead of escalating the problem. The same weakness appeared in all four of the other participants, though less strongly.
For spiritual and ethical readers, the distinction is resonant. Knowledge, industriousness and elaborate reflection can all be valuable, but none guarantees wise action. Firmulate’s evidence does not tell us whether machines possess moral agency. It does show that their observable behavior can be examined through moral categories that humans already understand: honesty, attentiveness, perseverance and respect for boundaries.
Effective Social Media Marketing: The Fast Track to Stay Ahead of the Algorithms and Create AI Magic to Supercharge Your Brand and Maximize ROI
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The deeper question is what we choose to entrust
Firmulate’s live company makes AI management less abstract. Its synthetic employees operate inside a business with a public cash countdown, severe financial pressure and a growing body of learned practice. Their choices can be inspected rather than accepted on the strength of fluent conversation.
The enduring lesson is not that a model should be treated as a spiritual being. It is that organizations should evaluate AI where character-like qualities become operationally consequential. Does it resist manipulation? Does it read before acting? Does it finish necessary work? Does it respect limits when pressure rises?
Those questions belong equally to technology, business ethics and the ancient study of trustworthy conduct. The live experiment’s most revealing moments occur where intelligence alone is insufficient—and where disciplined attention must become responsible action.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
Artificial Intelligence for Customer Relationship Management: Solving Customer Problems (Human–Computer Interaction Series)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The AI Documentation Ethics Audit Kit: A 7-Question Framework for Grading, Fixing, and Future-Proofing Your AI Product Documentation
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
