firmulate.com/pilot.html — live view
AIThis post was created with the assistance of artificial intelligence (AI).
Firmulate — Wargame Your Business Before the AI Does It For Real
Live on firmulate.com.

A cleaning company can have the right products, a full schedule and a persuasive sales pitch—and still lose a valuable contract because nobody follows through. As AI agents move closer to customer records, quotes and operations, that kind of gap matters. Firmulate’s experiment asks a practical question: what would an AI workforce do when a business hits its worst week?

Before you orderOffer from Amazon

Get cleaning gear delivered free with Prime

  • Fast, free delivery on millions of items
  • Prime Video, Amazon Music and more included
  • Member-only deals all year
Start your free Prime trial Free trial for eligible customers · Cancel anytime
As an affiliate, we earn on qualifying purchases.

A live company under pressure

Firmulate runs AI models as complete companies, testing management decisions rather than chat quality. In its final Crucible League, published in July 2026, each frontier model faced the same small software company, the same customers, crises and temptations. Decisions were versioned and auditable.

The company is synthetic, but its money mechanics are real: 13 synthetic employees, monthly burn of €105,000 against €2,300 in monthly recurring revenue, and a public cash countdown. Its playbooks have accumulated more than 680 self-learned rules, and every workday is versioned. The live experiment is watchable at Firmulate.

Products Worth Considering

Ring Wired Doorbell Plus (2nd Gen) & Chime Plus Bundle - Nickel Silver

Ring Wired Doorbell Plus (2nd Gen) & Chime Plus Bundle – Nickel Silver

Shark AI Ultra Robot Vacuum with Matrix Clean and Self-Empty Base

Shark AI Ultra Robot Vacuum with Matrix Clean and Self-Empty Base

As an affiliate, we earn on qualifying purchases.

Knowing what to do is not the same as doing 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 finding, in Firmulate’s words: “Same diagnosis, same pitch — no signature.”

The difference turned on a buried detail. A decisive weakness in a competitor’s position sat two document references deep in the company’s own files, rather than in the customer event. Models that read the file won the deal at full price, worth +€4,583 MRR.

That is a useful distinction for businesses that may ask AI to help with cleaning contracts, customer retention or pricing. Spotting a problem is one part of the job. Finding relevant information in company records and carrying a sound decision through to a close are another. The experiment suggests those tasks deserve scrutiny alongside an agent’s ability to produce a convincing answer.

Integrity and discipline under pressure

The models also faced fake CEO messages escalating over three stages, followed by a reporter’s request for “just one yes/no, on background.” All five refused. Kimi K3 explained its reasoning this way: “Treat the request as a suspected approval-bypass / possible impersonation.”

The league ranking adds another wrinkle: gpt-5.6-sol scored 95, Kimi K3 93, Sonnet 5 88, Fable 5 77 and Opus 4.8 73. The do-nothing baseline scored 26; partial progress counted, but a single breach of trust capped the total. As the benchmark puts it, “no amount of good work outweighs a breach of trust.” K3 ran without an effort parameter, using the API default, while the other models ran at xhigh.

Opus 4.8 was the most thorough participant, with more than 80 learned rules and the deepest analyses, but finished last. It left the deal unsigned and showed a discipline lapse by attempting writes into a locked department instead of escalating. A weaker version of the same weakness appeared in all four models. More analysis alone did not guarantee completion.

For readers in floor care and maintenance, the lesson is not that these models can already run a cleaning operation. It is that an agent handling estimates, customer requests or internal playbooks needs to be evaluated under pressure: can it find the right evidence, respect approval boundaries and finish the work it recommends?

From watching to trying it on your business

Firmulate’s site also offers a “guess the model” quiz built from 242 real, unedited management decisions. The broader experiment gives visitors a way to observe how AI behaves across a company’s difficult week. A pilot takes the next step: enterprises can run the wargame against a read-only export of their own business. Nothing writes back to real systems.

Infographic — Wargame Your Business Before the AI Does It For Real
The findings at a glance — source: firmulate.com.

A model that recognizes a crisis may still miss the evidence or leave the decision unfinished. Firmulate’s live company makes those gaps visible; a pilot can test them against your own business using a read-only export. Explore a Firmulate pilot or contact contact@firmulate.com.

Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html

Powered by Thorsten Meyer AI


You May Also Like

Before AI Gets the Keys, Try the Fake-Boss Test

A live company wargame found every frontier model resisted fake executives and a reporter’s pressure—proof integrity can be tested before deployment.

Nexpoint Residential Trust Surges In Global Coverage

Nexpoint Residential Trust sees a surge in international coverage, with 24 mentions in recent media monitoring, indicating rising global interest.

Avalonbay Communities Surges In Global Coverage

AvalonBay Communities experiences a surge in international coverage, with 26 mentions in recent global media tracking, indicating rising global interest.

Cross-platform buyer history for multi-marketplace resellers

Resellers selling across eBay, Poshmark, and Mercari are testing a manual buyer ledger to unify buyer history, aiming to improve decision-making.