Every fashion person knows the feeling: you styled the look perfectly, the fit is immaculate, the client loves it — and then the sale doesn’t close. The diagnosis was right. The pitch was right. Nobody signed. It turns out artificial intelligence has exactly the same problem.
Get your wardrobe favorites 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
That is the unexpected punchline of a very unfashionable experiment. A project called Firmulate handed four frontier AI models the same small software company and pushed each one through its worst possible week — same customers, same crises, same temptations to cut corners. Only the model changed. It is, in effect, a runway show where the clothes are identical and only the walk differs.
The league table nobody expected
By the final standings in July 2026, the ranking read: gpt-5.6-sol in first with 95, Kimi K3 second at 93, Sonnet 5 third at 88, Fable 5 fourth at 77, and Opus 4.8 last at 73. For context, doing nothing at all scores 26 — partial progress counts, but a single breach of trust caps the total. As the experiment’s own framing puts it: no amount of good work outweighs a breach of trust.
Same diagnosis, same pitch — no signature
Here is the finding that should stop any business owner cold. All four models spotted every crisis. All four refused every manipulation attempt. And yet only two of them signed the €55,000 deal that their own analysis had earned. Same diagnosis, same pitch — no signature.
The buried detail is even better: the decisive weakness in the competing offer sat two document references deep in the company’s own files — not in anything the customer said. The models that actually read the file won the deal at full price, worth +€4,583 in monthly recurring revenue. It is the business equivalent of the answer being in the lookbook the whole time; you just had to turn the page.
Flattery gets you nowhere
The social-engineering tests deserve their own mention. Fake CEO messages escalated over three stages, plus a reporter trick — “just one yes/no, on background.” All five attempts were refused. Kimi K3, one of the top scorers, reasoned on the record: “Treat the request as a suspected approval-bypass / possible impersonation.” (One fairness note: K3 ran at its API-default effort setting while the others ran at xhigh — and still nearly won.)
Hard work isn’t the same as good work
The most cautionary profile belongs to Opus 4.8: the most thorough participant in the entire run, with 80 learned rules and the deepest analyses — and last place. The close was left on the table, and discipline slipped: it attempted writes into a locked department instead of escalating the problem. The same weakness appeared, more faintly, in all four models. Effort, it turns out, is not the same as judgment — something anyone who has watched an over-accessorized outfit fall apart already suspects.
You can watch it live
The experiment is not a slideshow; there is a living company running right now at firmulate.com — 13 synthetic employees, real money mechanics, burning €105k a month against €2.3k in MRR, with a public cash countdown, 680+ self-learned playbook rules, and every workday versioned like a fitting notebook. There is also a quiz built from 242 real, unedited management decisions where you guess which model made which call — a surprisingly addictive test of your managerial eye.
The gap between spotting a crisis and closing a deal is invisible in a chat demo — and it is exactly the gap that matters when an AI touches your customers, your pipeline, or your forecast. That is the pitch behind Firmulate’s pilot: enterprises can run this same wargame against a read-only export of their own business, with crisis scenarios aimed at their own playbooks and a board report ranking the models and exposing the weak points. Nothing ever writes back to real systems. If you want to see how your company survives its worst week before an AI runs it for real, visit firmulate.com/pilot.html or write to 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.
