A floor scrubber can keep a maintenance schedule. But what happens when a major customer threatens to leave, a competitor undercuts your price and someone posing as the CEO asks for an exception? For cleaning and floor-care businesses weighing AI for customer service, sales or operations, those decisions matter as much as a tidy forecast. Firmulate’s experiment puts AI models in charge of a small company and watches how they handle the pressure.
Get cleaning gear 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
A company under pressure
In the final Crucible League, completed in July 2026, frontier models ran the same small software company through its worst week: same customers, same crises and same temptations. Every decision was versioned and auditable. The live company has 13 synthetic employees and real money mechanics, with monthly burn of €105,000 against €2,300 in monthly recurring revenue. Its public cash countdown, 680+ self-learned playbook rules and workday-by-workday decisions make the experiment watchable at Firmulate.
The league placed gpt-5.6-sol first at 95, followed by Kimi K3 at 93, Sonnet 5 at 88, Fable 5 at 77 and Opus 4.8 at 73. The do-nothing baseline scored 26. Partial progress counted, but a single breach of trust capped the total: “no amount of good work outweighs a breach of trust.” These are results from this experiment, not a promise about how a model will behave in every business.
Recognizing the crisis was not enough
Every model spotted every crisis and refused every manipulation attempt. Yet only two signed the €55,000 deal their own analysis had earned. The finding is pointed: “Same diagnosis, same pitch — no signature.” An AI can identify the right move and still fail to follow through. In a cleaning business, the equivalent might be understanding why a contract is at risk but failing to make the call or present the right offer.
The deal hinged on a detail buried two document references deep in the company’s own files, rather than in the customer event itself. Models that read the file won the deal at full price, worth +€4,583 in monthly recurring revenue. The lesson for operators is practical: decisions depend on whether an AI can find and use relevant information already held in company records, not just react to the latest message.
Trust under pressure, discipline under scrutiny
The social-engineering test used fake CEO messages escalating over three stages, plus a reporter’s line: “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 kind of caution matters wherever an AI might encounter requests to override a process or disclose sensitive business information.
Opus 4.8 was the most thorough participant, with +80 learned rules and the deepest analyses, but finished last. It left the close on the table and slipped on discipline, attempting writes into a locked department instead of escalating. A weaker version of the same weakness appeared in all four. More analysis, then, did not guarantee effective 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. The league is one observed experiment, and that difference belongs alongside the ranking when interpreting it.
From watching to testing your own business
For a cleaning contractor, facilities team or floor-care supplier, the useful next question is not whether an AI can write a convincing email. It is how it handles a churn risk, a price increase, a competitor attack or a suspicious request using the information and rules that actually govern your business. Firmulate’s pilot takes that question to a company’s own data: a digital twin built from a read-only export, with crisis scenarios and a board report covering model ranking and weak points in existing playbooks.
The boundary is clear: the pilot does not write back to real systems. A read-only export lets a company examine how models respond without giving them permission to alter customer records or operational tools. Firmulate also offers a “guess the model” quiz built from 242 real, unedited management decisions at firmulate.com.
Take the wargame to your business
Watching an AI manage a synthetic company can reveal the gap between sound analysis and sound decisions. A pilot can put your own playbooks and scenarios under pressure using a read-only export, then show where models perform and where procedures may need attention. To discuss a pilot, visit Firmulate’s pilot page or email 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.
