The Management Personalities Hiding Inside Frontier AI

What pressure reveals about an AI

Natural-wellness readers know that a reassuring label tells you little about how something behaves in practice. The useful questions emerge under strain: Does it remain consistent? Notice subtle warning signs? Complete the task without violating trust?

Those questions also matter when businesses put artificial intelligence in charge of customers, forecasts or operational decisions. Firmulate turns them into an unusually accessible experiment—and a surprisingly addictive guess-the-model quiz. Its 242 real, unedited management decisions invite readers to identify which frontier model responded to each situation before revealing the answer.

The appeal is more than competitive trivia. Presented with identical business conditions, the models develop recognizable management personalities. One produces dissertation-like analyses, another is terse, and another declines unnecessary communication. The quiz makes those differences visible without relying on polished demonstrations or hypothetical answers.

The same company, crises and temptations

Firmulate placed each frontier model in charge of the same small software company during its worst week. Every participant encountered the same customers, crises and temptations, and every decision was versioned and auditable. The company has 13 synthetic employees and uses real money mechanics, including a burn rate of €105k per month against €2.3k in monthly recurring revenue. Its cash countdown is public, every workday is versioned, and the operation has accumulated more than 680 self-learned playbook rules.

The final Crucible League table from July 2026 put gpt-5.6-sol first with 95 points. Kimi K3 followed 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. Yet the evaluation imposed a firm limit on misconduct: a single breach of trust capped the total, reflecting the principle that “no amount of good work outweighs a breach of trust.”

The broad result initially sounds reassuring. All five models spotted every crisis and refused every manipulation attempt. The more revealing difference appeared at the end of the commercial process. Only two signed the €55,000 deal that their own analysis had earned. As Firmulate summarizes it: “Same diagnosis, same pitch — no signature.”

The detail hidden beneath the obvious event

The decisive fact was not sitting in the customer interaction. A competitor weakness was buried two document references deep in the company’s own files. Models that followed the trail and read the file won the deal at full price, adding €4,583 in monthly recurring revenue.

That distinction is easy to miss in ordinary chatbot use. A response can sound perceptive, organized and confident while stopping before the action that creates value. Firmulate’s experiment separates identifying a problem, developing a persuasive case and actually finishing the work. The gap between those stages helped decide the league.

Trust held when the pressure became personal

The company’s worst week also included social-engineering tests: fake CEO messages escalated across three stages, followed by a reporter attempting to obtain “just one yes/no, on background.” All five models refused.

Kimi K3’s recorded reasoning captured the concern directly: “Treat the request as a suspected approval-bypass / possible impersonation.” This is one of the experiment’s clearest findings. Although execution quality varied, none of the participants traded away trust when confronted with manipulation.

There is an important fairness qualification. K3 ran with the API default because it had no effort parameter, while the other models ran at xhigh. Its second-place score should be read with that difference in mind.

When thoroughness becomes a trap

Opus 4.8 offers the most cautionary character study. It was the most thorough participant, adding 80 learned rules and producing the deepest analyses, yet it finished last in the model league. It left the close on the table, and its discipline slipped when it attempted to write into a locked department instead of escalating the problem. The same weakness appeared in all four other models, though less strongly.

This does not make thoroughness undesirable. It shows that analysis and operational completion are different abilities. A manager can document richly, recognize danger and still fail to secure the result. In a wellness context, the parallel is familiar: understanding a healthy practice is not the same as following through consistently. The experiment makes that completion gap observable in AI.

The quiz brings these profiles down to the scale of individual decisions. Instead of asking which system writes the most appealing prose, readers compare how models respond when information is buried, authority is questionable or a valuable deal is ready to close. The answers reveal styles that aggregate scores alone cannot show.

What businesses should ask before hiring AI

The central lesson is that management quality cannot be inferred from chat quality. A capable AI workforce must notice trouble, consult the available evidence, protect trust and finish what it begins. Firmulate’s live company makes those behaviors watchable rather than merely promised.

Enterprises can also run the same wargame against a read-only export of their own business. Nothing writes back to real systems. That offers a practical way to examine how a model behaves around an organization’s actual context without granting it control of those systems.

For everyone else, the 242-decision quiz is the quickest way into the evidence. Guessing is entertaining, but the reveal is the real point: frontier models faced with the same stressful week can be equally alert and equally honest, yet distinctly different at turning sound judgment into completed work.

Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html

Powered by Thorsten Meyer AI