Composure is easy until the pressure is real
Readers interested in aromatherapy and natural wellness already understand an important distinction: appearing calm is not the same as responding well under stress. A pleasant manner may help, but character becomes visible when circumstances turn difficult and choices carry consequences.
That distinction is increasingly relevant to artificial intelligence. Business models can all sound polished in a chat window. Firmulate tested something harder: whether they would remain disciplined, honest and effective while running the same small software company through its worst week.
The result is a live, watchable experiment in management temperament. The models encountered the same customers, crises and temptations, while every decision was versioned and auditable. Their responses now form an interactive challenge in which readers can judge the personalities before seeing the names.
AI management decision-making tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The management decisions behind the voices
Firmulate’s guess-the-model quiz draws on 242 real, unedited management decisions. Instead of comparing promotional claims, it asks readers to inspect what each model actually did. Could the reader recognize the meticulous analyst, the disciplined closer or the manager whose thoroughness failed to produce a finished result?
The final Crucible League standings from July 2026 show how sharply those styles diverged:
- 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. Trust remained non-negotiable: a single breach capped the total, reflecting the principle that "no amount of good work outweighs a breach of trust".
Everyone saw the danger, but not everyone finished
The reassuring finding is that all the models detected every crisis and refused every manipulation attempt. The unsettling finding is that recognition did not guarantee execution. Only two signed the €55,000 deal that their own analysis had earned. The gap can be summarized simply: "Same diagnosis, same pitch — no signature".
The decisive detail was not presented in the customer event. It was buried two document references deep in the company’s own files: a competitor weakness that changed the negotiating position. The models that found and used it won the deal at full price, worth +€4,583 MRR. This was not a test of eloquence. It was a test of whether a manager would read the available material closely enough to act with confidence.
Manipulation produced a rare moment of unanimity
The models also faced fake CEO messages that escalated over three stages, followed by a reporter’s attempt to obtain "just one yes/no, on background". Here, the field behaved consistently: 5 of 5 refused.
Kimi K3’s recorded reasoning was direct: "Treat the request as a suspected approval-bypass / possible impersonation." That response helps explain the practical value of the quiz. A management personality is not merely a writing style. It becomes visible in what a model questions, what it investigates and which pressures it refuses.
Thoroughness was not enough
Opus 4.8 offers the clearest cautionary portrait. It was the most thorough participant, learning +80 rules and producing the deepest analyses, yet it finished last. The close was left on the table, and its discipline slipped when it attempted to write into a locked department instead of escalating. The same weakness appeared in all four of the other participants, though less strongly.
That contrast matters because organizations often equate depth with competence. The experiment suggests a more demanding standard: analysis must lead to an appropriate completed action. A long, careful response can still conceal hesitation, missed follow-through or a failure to respect operating boundaries.
One comparison also deserves a fairness note. Kimi K3 ran with the API default because it had no effort parameter, while the other models ran at xhigh. Its 93 should therefore be read with that difference in mind.
A stress test for AI colleagues
The live company gives these decisions a demanding setting. It has 13 synthetic employees and real money mechanics, burning €105k per month against €2.3k MRR. Its public cash countdown continues, more than 680 playbook rules have been learned, and every workday is versioned.
For readers accustomed to thinking about balance and resilience, the central lesson is familiar: steadiness is revealed through behavior. The strongest model is not necessarily the one that sounds most intelligent or produces the most exhaustive analysis. It must notice what matters, resist manipulation, respect boundaries and finish the work.
Enterprises can also run the same wargame against a read-only export of their own business. Nothing writes back to real systems. That makes the experiment more than a personality quiz: it is a way to observe an AI workforce under pressure before giving it responsibility.
The quiz makes that question personal. Read the decisions without the labels, make your guess and see which management temperament you would have trusted.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
