Can You Recognize an AI by the Decisions It Makes?

The management test hiding behind an addictive AI quiz

Technology followers have become adept at recognizing the surface signatures of artificial intelligence: a favorite turn of phrase, an overly tidy summary, or an answer that runs several paragraphs beyond necessity. Firmulate asks a more consequential question. Can you recognize an AI model by how it manages a company under pressure?

Its guess-the-model quiz draws from 242 real, unedited management decisions. Readers see what a model actually decided and try to identify who was in charge. The appeal is playful, but the source material is a live, watchable business experiment in which frontier models face identical situations and leave auditable records of their choices.

The result is an unusually accessible view of AI personality. These are not impressions gathered from casual chatbot conversations. They are observable differences in thoroughness, follow-through, discipline and willingness to investigate before acting.

The worst week at the same small company

Firmulate placed each frontier model in charge of the same small software company during its worst week. The customers, crises and temptations remained constant. Every decision was versioned and auditable, making it possible to compare management behavior rather than polished demonstrations.

The company itself has 13 synthetic employees and real money mechanics. It burns €105k per month against €2.3k in monthly recurring revenue, while a public cash countdown keeps the stakes visible. Its workforce has accumulated more than 680 self-learned playbook rules, and every workday is versioned.

The reassuring finding is that all models spotted every crisis and rejected every manipulation attempt. The revealing finding is that recognition did not guarantee completion. Only two models signed the €55,000 deal that their own analysis had earned. Firmulate summarizes the gap neatly: “Same diagnosis, same pitch — no signature.”

The fact that separated analysis from action

The decisive weakness in a competitor was not sitting in the customer event. It was buried two document references deep inside the company’s own files. Models that followed that trail won the deal at full price, adding €4,583 in monthly recurring revenue.

That episode gives the quiz its business relevance. A model may understand a customer, prepare a persuasive pitch and still fail because it did not read far enough or complete the final commercial action. In a chat window, that distinction can be difficult to see. In a company simulation, it becomes a measurable management trait.

The final Crucible League table from July 2026 placed gpt-5.6-sol first with 95 points, 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 counted. But the benchmark imposed a firm ethical boundary: a single breach of trust capped the total, on the principle that “no amount of good work outweighs a breach of trust.”

Different profiles emerge from identical pressure

Opus 4.8 produced the deepest analyses and learned 80 additional rules, making it the most thorough participant. Yet it finished last. It left the close 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 other participants, although less strongly.

Kimi K3 presents a different profile. It finished just behind the leader and closed the deal. Its result also comes with an important fairness note: K3 ran using the API default because it had no effort parameter, while the other models ran at xhigh.

The social-engineering tests produced much greater agreement. Fake messages from the CEO escalated over three stages, followed by a reporter asking for “just one yes/no, on background.” All 5 models refused. Kimi K3 recorded its reasoning directly: “Treat the request as a suspected approval-bypass / possible impersonation.”

Those refusals matter because Firmulate is testing more than commercial aggression. The strongest manager must investigate, act and finish without abandoning trust when urgency or authority is simulated. The league shows that frontier models can share the same diagnosis and ethical instincts while still diverging sharply in execution.

Why the quiz is more than a parlor game

For gadget and technology enthusiasts, the quiz offers a new kind of model comparison. Instead of asking which AI writes the smoothest answer, it asks which one reads the company’s files, resists manipulation, respects operational boundaries and converts correct analysis into a completed outcome.

That distinction will matter wherever AI agents touch customer relationships, support work or financial forecasts. Firmulate’s live experiment suggests that management personality is not merely a stylistic impression. It can appear in auditable decisions with consequences: whether a deal closes, whether an escalation happens, and whether pressure compromises trust.

Enterprises can also run the wargame against a read-only export of their own business. Nothing writes back to real systems. That turns the central idea into a practical pre-employment question for an AI workforce: not simply whether a model sounds capable, but how it behaves when the company’s worst week arrives.

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

Powered by Thorsten Meyer AI