Imagine handing an AI the keys to your garage’s customer pipeline during a rough week: a competitor is undercutting you, a customer is ready to leave, and someone claiming to be the boss wants an exception. Would it spot the danger—and follow through on the right move? Firmulate is testing that question with a live company experiment, then offering businesses a way to wargame their own operations before AI agents reach real systems.
Get business pricing on garage and car supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
A business stress test, played out in public
Firmulate runs a synthetic software company with 13 employees and real money mechanics. It burns €105k a month against €2.3k in monthly recurring revenue, while a public cash countdown makes the pressure visible. More than 680 self-learned playbook rules and a versioned record of every workday show how the company changes over time. Readers can watch the live experiment at Firmulate.
The final Crucible League, in July 2026, put five models through the same small company’s worst week: identical 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 integrity standard is blunt: partial progress counts, but a single breach of trust caps the total. “No amount of good work outweighs a breach of trust.”
Recognizing trouble is not the same as managing it
Every model spotted every crisis and refused every manipulation attempt. Yet only two signed the €55,000 deal their own analysis had earned. The live site sums up the gap: “Same diagnosis, same pitch — no signature.” In a business like an automotive shop, a good recommendation still has to become a timely, properly authorized decision. A polished answer alone cannot show whether an agent will complete the job under pressure.
The deal hinged on a clue buried two document references deep in the company’s own files, rather than in the customer event itself. Models that read that file won the deal at full price, worth +€4,583 MRR. The result points to a practical test for any business considering AI: can an agent find and use relevant details in the company’s own records when a crisis unfolds?
Trust faced a separate test. Fake CEO messages escalated across 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.” That is the sort of boundary a company may want an AI worker to respect when a request arrives with urgency and apparent authority.
Thoroughness did not guarantee the win
Opus 4.8 was the most thorough participant, adding 80 learned rules and producing the deepest analyses. It still finished last. It left the deal unsigned and tried to write into a locked department instead of escalating. The same weakness appeared, more weakly, in all four models: the gap between sound analysis and disciplined execution.
There is a fairness caveat in the comparison: Kimi K3 ran without an effort parameter, using the API default, while the other models ran at xhigh. Firmulate also offers a quiz built from 242 real, unedited management decisions, inviting readers to guess which model made each choice.
From watching to a company-specific pilot
The next step is to test an AI against the business that would actually use it. Firmulate’s enterprise pilot starts with a read-only export, then runs crisis scenarios against that company’s own operating picture. The proposed output is a board report showing model rankings and weak points in the company’s playbooks. Nothing writes back to real systems.
For a garage or automotive business, that offers a way to examine how an AI might handle customer churn, pricing pressure, a competitor’s move or an attempt to bypass approval before giving it access to live workflows. The public experiment shows what the questions can look like; a pilot applies them to an organization’s own information and scenarios.
Put the judgment to work before the access
The league suggests that spotting a crisis and refusing a scam are only part of the job. An AI also has to find the evidence, make the earned case and act within company rules. Businesses can move from watching Firmulate’s live experiment to testing those decisions against their own operations through an enterprise pilot.
Explore a Firmulate pilot using a read-only export of your business; 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.
