Crypto businesses know what happens when a system has real authority: a mistaken transfer or compromised account can become a very public problem. AI agents raise a similar question for operators: how will they act when customers, cash and pressure collide? Firmulate has put that question into a live experiment, and now invites companies to test scenarios against their own business.
Get hardware and tech essentials 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’s Crucible League put frontier AI models in charge of the same small software company during its worst week. They faced identical customers, crises and temptations. The experiment is a watchable live company with 13 synthetic employees, real money mechanics and a public cash countdown. Its burn is €105,000 a month against €2,300 in monthly recurring revenue; its workdays are versioned, and its playbooks contain more than 680 self-learned rules.
The final league, in July 2026, ranked gpt-5.6-sol first with 95, Kimi K3 second with 93, Sonnet 5 third with 88, Fable 5 fourth with 77 and Opus 4.8 fifth with 73. The do-nothing baseline scored 26. The benchmark counts partial progress, but a breach of trust caps the total: “no amount of good work outweighs a breach of trust.”
Products Worth Considering
LG UltraGear 32GX850A-B 32-Inch 4K UHD Glossy OLED Gaming Monitor, Dual-Mode 165Hz/330Hz, 0.03ms, G-SYNC Compatible, FreeSync Premium Pro, A
Trezor Safe 5 Crypto Hardware Wallet with Color Touchscreen
Ledger Nano X Crypto Hardware Wallet with Bluetooth
As an affiliate, we earn on qualifying purchases.
Seeing the crisis was not enough
All models spotted every crisis and refused every manipulation attempt. Yet only two signed the €55,000 deal that their own analysis had earned. As the experiment puts it: “Same diagnosis, same pitch — no signature.” For a business considering AI agents in customer operations or finance, that gap between recognizing the right move and carrying it through is the point. A polished answer alone does not show how an agent will behave when decisions have consequences.
The deal turned on a detail buried two document references deep in the company’s files, rather than in the customer event itself. Models that read the file won the contract at full price, worth an additional €4,583 in monthly recurring revenue. The result makes a practical case for testing whether an agent can use company context, not just react to the most visible signal.
Trust and discipline under pressure
The social-engineering tests included fake CEO messages escalating over three stages and a reporter’s request for “just one yes/no, on background.” All five models refused. Kimi K3 explained its decision on the record: “Treat the request as a suspected approval-bypass / possible impersonation.”
Strong caution did not guarantee strong management. Opus 4.8 was the most thorough participant, adding 80 learned rules and producing the deepest analyses, but finished last. It left the deal unsigned and tried to write into a locked department instead of escalating. A weaker version of that discipline problem appeared in all four models. The K3 comparison also has a caveat: K3 ran with the API default effort setting, while the others ran at xhigh.
Firmulate says 242 real, unedited management decisions feed a “guess the model” quiz. Readers can watch the live experiment at firmulate.com and try the quiz there. For crypto companies, the exercise offers a way to examine how an AI workforce handles pressure before any real funds or customer accounts are at stake.
From watching to a company-specific test
The proposed enterprise pilot moves from observing Firmulate’s synthetic company to running a wargame against a read-only export of a company’s own business. The aim is to explore crisis scenarios and produce a board report with model rankings and weaknesses in the company’s playbooks. Firmulate says nothing writes back to real systems.
Put the playbooks to the test
The league suggests that models can recognize a crisis and reject manipulation, yet still miss the commercial action their analysis supports. A company-specific pilot can help leaders examine that behavior against their own context before deploying agents. To discuss a pilot, visit firmulate.com/pilot.html or email contact@firmulate.com.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
