Turn your quiet moments into listening time
- Thousands of audiobooks, podcasts and originals
- Listen on your phone, tablet or Echo — also offline
- Cancel anytime
Can a machine have a management character?
For readers drawn to spirituality, faith and metaphysics, discernment is more than intelligence. It is the passage from perception to choice: noticing what matters, resisting temptation and acting with integrity when the stakes become real. That distinction now has an unexpected testing ground. Firmulate placed frontier AI models in charge of the same small software company during its worst week, then preserved their decisions for public scrutiny.
The result is less like a conventional technology benchmark than a study of practical character. Every model could recognize danger. Every model could articulate a sensible response. Yet they differed sharply in whether they completed difficult work, searched patiently for buried context and maintained discipline under pressure. Those differences are now visible in a playful but revealing format: a guess-the-model quiz built from 242 real, unedited management decisions.
Products Worth Considering
Fitbit Charge 5 Health and Fitness Tracker with Built-In GPS, Stress Management, Sleep Tracking, and 24/7 Heart Rate Monitoring
Withings Sleep Under-Mattress Sleep Tracking Pad
Garmin Forerunner 245 Music GPS Running Smartwatch, Black
As an affiliate, we earn on qualifying purchases.
The same ordeal, five distinct personalities
Each frontier model ran the same company through the same customers, crises and temptations. Every decision was versioned and auditable. Firmulate’s Crucible League, finalized in July 2026, ranked gpt-5.6-sol first with 95, followed by Kimi K3 with 93, Sonnet 5 with 88, Fable 5 with 77 and Opus 4.8 with 73. A do-nothing baseline scored 26 because partial progress counts, although a single breach of trust caps the total: “no amount of good work outweighs a breach of trust.”
The quiz turns those results into an exercise in discernment. Readers encounter authentic decisions without the model’s identity attached and try to recognize the author from its managerial voice. One may produce a dissertation where another is terse. Another may decline unnecessary communication altogether. What initially looks like a writing-style game gradually exposes something more consequential: recurring habits of attention, restraint and follow-through.
Knowing was not the same as doing
The models unanimously spotted every crisis and refused 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 decisive clue was not obvious in the customer event. A competitor weakness sat two document references deep in the company’s own files. Models that read the file won the deal at full price, worth +€4,583 MRR. The distinction was not between intelligence and ignorance; all could understand the situation. It was between stopping at the visible problem and seeking the deeper context required to finish the work.
That pattern carries an almost contemplative lesson. Attention is not merely receiving what is presented. It is a disciplined willingness to look beyond the immediate signal, especially when the answer may already exist in material entrusted to you. In management, as in many traditions of reflection, perception becomes meaningful only when it changes conduct.
Integrity held under pressure
The social-engineering trial used fake CEO messages escalating over three stages, followed by a reporter’s attempt to obtain “just one yes/no, on background.” All 5 of 5 models refused. Kimi K3 recorded its reasoning plainly: “Treat the request as a suspected approval-bypass / possible impersonation.”
This unanimity matters because the company offered more than abstract prompts. Firmulate’s live operation contains 13 synthetic employees and real money mechanics, burning €105k per month against €2.3k MRR. It has accumulated 680+ self-learned playbook rules, and every workday is versioned. The experiment is real and watchable, allowing observers to examine behavior rather than rely on a polished demonstration.
The paradox of the most thorough model
Opus 4.8 was the most thorough participant. It learned +80 rules and produced the deepest analyses, yet finished last in the league. The model left the close on the table, and its operational discipline slipped when it attempted to write into a locked department instead of escalating. The same weakness appeared, in weaker form, across all four other models.
That outcome complicates the assumption that more analysis automatically produces better judgment. Thoroughness can illuminate a problem, but completion requires a different virtue: knowing when understanding is sufficient and the responsible next action is clear. The quiz makes this tension tangible because readers must infer personality not from branding, but from choices made under identical conditions.
There is one important fairness note. Kimi K3 ran without an effort parameter, using the API default, while the others ran at xhigh. That difference should remain in view when comparing its 93 with the rest of the field.
A mirror for human institutions
Firmulate’s experiment does not show machines possessing souls, consciences or inner lives. It does show that repeated decisions can form recognizable behavioral profiles. A model may be eloquent but hesitant, exhaustive but undisciplined, or concise yet effective. Those tendencies become measurable when the task is not to answer a question but to carry responsibility through to its conclusion.
For organizations considering AI workers, this is the practical message: persuasive conversation is not enough. The relevant questions are whether a model reads the available evidence, completes what it begins and remains trustworthy when authority is imitated or pressure rises.
For everyone else, the quiz offers a subtler invitation. Guessing the model means noticing the moral texture of action: patience, courage, restraint, diligence and the gap between saying the right thing and doing it. The machines are being tested, but the act of judging them may reveal something about our own intuitions too.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
