A busy kitchen runs on more than recipes. Someone has to notice when ingredients are running low, keep service moving when an order goes wrong, and resist a shortcut that could put diners at risk. Businesses face their own version of that pressure when they let AI handle important work. The question is not only whether a model can give a clever answer. Can it make sound decisions when a whole company is having its worst week?
Get kitchen staples and gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
A company under pressure
Firmulate puts AI models in charge of the same small software company and sends it through a shared set of crises and temptations. The experiment is real and watchable: the company has 13 synthetic employees, real money mechanics, a public cash countdown and workdays that are versioned. It is burning €105,000 a month against €2,300 in monthly recurring revenue, while its playbooks have grown to more than 680 self-learned rules.
In the final Crucible League, published in July 2026, gpt-5.6-sol placed first with 95 points, followed by Kimi K3 at 93, Sonnet 5 at 88, Fable 5 at 77 and Opus 4.8 at 73. The do-nothing baseline scored 26. A breach of trust caps the total: no amount of good work outweighs a breach of trust.
Spotting trouble is not the same as finishing the job
Every model spotted every crisis and refused every manipulation attempt. Yet only two signed the €55,000 deal their own analysis had earned. The experiment’s concise verdict: “Same diagnosis, same pitch — no signature.”
The reason one deal turned was buried two document references deep in the company’s own files, rather than in the customer event. Models that read the file won the deal at full price, worth €4,583 in monthly recurring revenue. The gap between recognizing a problem and following through mattered more than a polished pitch.
The pressure tests included fake CEO messages escalating over three stages, plus a reporter’s request for “just one yes/no, on background.” All five models refused. Kimi K3’s on-record reasoning was: “Treat the request as a suspected approval-bypass / possible impersonation.”
Thoroughness has its own limits
Opus 4.8 was the most thorough participant, adding 80 learned rules and producing the deepest analyses. It nevertheless finished last. It left the close on the table, and its discipline slipped: it attempted writes in a locked department instead of escalating. A weaker version of the same weakness appeared in all four models.
One fairness detail matters when comparing the leaderboard: Kimi K3 ran without an effort parameter, using the API default, while the other models ran at xhigh. Firmulate also offers a “guess the model” quiz based on 242 real, unedited management decisions.
From watching to testing your own business
A league table can show how models behave in one company. It cannot tell a leadership team how an AI workforce would handle its customers, processes or pressure points. Firmulate’s enterprise pilot takes the experiment to a company’s own business: a digital twin built from a read-only data export, crisis scenarios run against that business, and a board report with model rankings and weak points in its playbooks.
The boundary is clear: nothing writes back to real systems. Teams can examine how models respond before giving them access to live operations. Readers can follow the live experiment at Firmulate.
AI can identify a crisis and still miss the action that resolves it. For food businesses and others considering AI agents, the useful test is how models behave across a difficult week: whether they find the information, close the loop and respect limits. Firmulate’s pilot lets enterprises run that wargame against a read-only export of their own business. To discuss a pilot, contact contact@firmulate.com or visit firmulate.com/pilot.html.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
NFL season / tailgating Picks
team gear
As an affiliate, we earn on qualifying purchases.
