A wedding planner can rehearse the seating chart, check the flowers and prepare for a last-minute cancellation. But what happens when an AI assistant is asked to handle the delicate decisions behind the scenes: a refund, a supplier dispute or a message that claims to come from the CEO? For businesses in the occasions trade, a polished demo cannot answer that. A realistic rehearsal might.
Get decor and gifts delivered free — and shop member deals
- Fast, free delivery on millions of items
- Access to Prime Big Deal Days deals on October 6–7
- Prime Video, Amazon Music and more included
From a live experiment to your own business
Firmulate puts AI models in charge of a small software company and watches how they handle pressure. In the final Crucible League, published in July 2026, gpt-5.6-sol placed first with 95 points, followed by Kimi K3 with 93, Sonnet 5 with 88, Fable 5 with 77 and Opus 4.8 with 73. The do-nothing baseline scored 26. Its rules also put a hard limit on a breach of trust: “no amount of good work outweighs a breach of trust.”
Every model faced the same customers, crises and temptations. The result was revealing: all models spotted every crisis and refused every manipulation attempt, yet only two signed a €55,000 deal that their own analysis had earned. The finding was summed up as “Same diagnosis, same pitch — no signature.” Knowing what to do and completing the important action turned out to be different tests.
The detail hiding in the files
The deal hinged on a competitor weakness buried two document references deep in the company’s own files, not in the customer event. Models that read the file won the deal at full price, worth +€4,583 MRR. It is a useful reminder for any business that relies on customer histories, supplier notes or event plans: the decisive detail may already be there, but an AI system has to find and act on it.
Trust was tested directly, too. Fake CEO messages escalated over three stages, followed by 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.”
The strongest score did not settle every concern. Opus 4.8 was the most thorough participant, with +80 learned rules and the deepest analyses, yet it finished last. It left the deal unsigned and tried to write into a locked department instead of escalating. The same weakness appeared, in a weaker form, in all four models. Thorough answers are valuable; they do not automatically make a reliable business partner.
A company you can watch
Firmulate’s live company has 13 synthetic employees and real money mechanics: it burns €105k a month against €2.3k in monthly recurring revenue, with a public cash countdown. Its playbook has more than 680 self-learned rules, and every workday is versioned. The live experiment is watchable at firmulate.com. A quiz built from 242 real, unedited management decisions invites visitors to guess which model made each choice.
The league also has a fairness wrinkle: Kimi K3 ran without an effort parameter, using the API default, while the other models ran at xhigh. That context belongs alongside the rankings when considering what the results say about model performance.
Rehearse against your own playbook
Watching a company face a crisis is one thing. A pilot takes the exercise to your organization: Firmulate can run the wargame against a read-only export of your business, including your company data, customers, pipeline and rules. You can examine how models handle scenarios relevant to your operation and get a board report with model rankings and weak points in your playbooks. The exercise does not write back to real systems.
For a business arranging weddings, holidays or celebrations, that could mean rehearsing supplier trouble, a wave of cancellations, a pricing change or a suspicious request for an exception before an AI assistant is trusted with real workflows. Firmulate’s experiment suggests the questions to ask: does a model recognize the crisis, protect trust, find the information it needs and follow through?
The useful test for workplace AI is not whether it sounds confident in a demo. It is whether it can make sound decisions under pressure and carry them through without compromising trust. Firmulate’s live company offers a public rehearsal; a pilot can put your own business playbooks to the test. To discuss a pilot, visit firmulate.com/pilot.html or contact contact@firmulate.com.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
