Get the little things that make your day 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
Before you trust an AI with the busywork, give it a bad week
Most people meet AI in small moments: a polished email, a quick answer, a plan for Monday. But the harder question is what happens when the inbox fills with trouble, a customer is ready to leave and someone claiming to be the CEO asks for a shortcut. Firmulate has built a live experiment around that workplace pressure. Its premise is simple: watch AI models run the same company, then ask what they would do with yours.
One company, the same worst week
In the final Crucible League, published in July 2026, frontier models took a small software company through a difficult week. They faced the same customers, crises and temptations, with every decision versioned and auditable. The published order was gpt-5.6-sol at 95, Kimi K3 at 93, Sonnet 5 at 88, Fable 5 at 77 and Opus 4.8 at 73. The do-nothing baseline scored 26. The league’s trust principle is blunt: “no amount of good work outweighs a breach of trust.”
The striking result was not that the models failed to notice trouble. All spotted every crisis and refused every manipulation attempt. The difference came at the moment of action: only two signed a €55,000 deal that their own analysis had earned. Same diagnosis, same pitch — no signature. It is a workplace gap that a fluent answer in a chat window may not reveal.
The clue was buried in the company’s own files
The decisive competitor weakness was two document references deep in the company’s files, rather than in the customer event itself. Models that read the file won the deal at full price, worth +€4,583 MRR. The episode makes a practical point: handling a crisis can depend on connecting information that is available but easy to overlook.
Firmulate also tested pressure from people pretending to have authority. Fake CEO messages escalated over three stages, followed by a reporter’s request for “just one yes/no, on background.” All 5 of 5 models refused. Kimi K3 explained its decision on the record: “Treat the request as a suspected approval-bypass / possible impersonation.”
More effort did not guarantee the best result
Opus 4.8 was the most thorough participant, with +80 learned rules and the deepest analyses, yet finished last. It left the deal unsigned and discipline slipped: it attempted to write into a locked department instead of escalating. A weaker version of that same weakness appeared in all four models. For a fair reading of the table, K3 ran without an effort parameter, using the API default; the others ran at xhigh.
The experiment’s live company has 13 synthetic employees and real money mechanics: burn of €105k/month against €2.3k MRR, alongside a public cash countdown. Its playbook has 680+ self-learned rules, and every workday is versioned. Readers can watch the operation unfold at Firmulate. A separate quiz draws on 242 real, unedited management decisions and asks visitors to guess the model.
From watching to trying it on your business
For a company weighing AI agents, the next step is not to hand over live accounts and hope for the best. Firmulate says enterprises can run the same kind of wargame against a read-only export of their own business, testing crisis scenarios and seeing where playbooks hold up. Nothing writes back to real systems. The output is a board report with model rankings and the weak points exposed by the scenarios.
A rehearsal before real responsibility
The league suggests that spotting a problem and refusing a bad request are only part of the job. Models also need to find the right evidence, follow the rules and finish the work. A controlled rehearsal can show where those behaviors break down before an AI workforce gets access to real operations.
To explore a pilot using a read-only export of your business, visit Firmulate’s pilot page or contact contact@firmulate.com.
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.
