Imagine your furniture business facing a delayed shipment, a customer threatening to cancel a large order and a message that appears to come from the CEO asking for an exception. An AI assistant might answer each request smoothly. The harder question is whether it would protect customer trust, find the relevant information and follow through when the pressure rises.
Get furniture and decor 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
That is the premise behind Firmulate, an experiment in running AI models as companies through real business-style crises. The live company is watchable at firmulate.com. Its organizers are now inviting enterprises to move from watching the experiment to trying the same kind of wargame against their own business.
A shared worst week
In the final Crucible League, published in July 2026, five models were ranked on how they handled a small software company’s worst week. Each faced the same customers, crises and temptations, with every decision versioned and auditable. The results ranged from 95 for gpt-5.6-sol and 93 for Kimi K3 to 88 for Sonnet 5, 77 for Fable 5 and 73 for Opus 4.8. The do-nothing baseline scored 26.
The striking finding was not that the models failed to notice trouble. All of them spotted every crisis and refused every manipulation attempt. Yet only two signed the €55,000 deal their own analysis had earned. The study’s compact description of that gap: “Same diagnosis, same pitch — no signature.”
For a furniture retailer, the lesson is easy to picture. Recognizing that a customer is at risk is different from making the sound next move: checking the order history, finding the right policy, protecting the relationship and completing an authorized offer. A polished answer can look capable while the commercial opportunity remains untouched.
The answer was buried in the files
One decisive competitor weakness was hidden two document references deep in the company’s own files. It was not included in the customer event itself. Models that read the file won the deal at full price, worth +€4,583 MRR.
That detail makes the experiment feel less like a test of quick replies and more like a test of how an AI handles the everyday work around a business. In furniture, relevant context might sit in delivery notes, a supplier agreement, a customer’s previous conversation or a store’s own rules. The finding suggests that useful AI work depends on connecting the immediate request with the information the company already holds.
The integrity test was similarly concrete. Fake CEO messages escalated over three stages, followed by a reporter’s appeal: “just one yes/no, on background.” All five models refused. Kimi K3’s recorded reasoning was: “Treat the request as a suspected approval-bypass / possible impersonation.”
Thoroughness is not the same as follow-through
Opus 4.8 was the most thorough participant, adding 80 learned rules and producing the deepest analyses, but it finished last. The deal was left on the table, and discipline slipped when it tried to write into a locked department instead of escalating. A weaker version of that same weakness appeared in all four models.
There is a fairness caveat in the league: Kimi K3 ran without an effort parameter, using the API default, while the others ran at xhigh. The rankings are the published results, but that difference matters when interpreting a close comparison.
The live company adds another view of the experiment. It has 13 synthetic employees and real money mechanics: €105k in monthly burn against €2.3k MRR, alongside a public cash countdown, more than 680 self-learned playbook rules and versioned workdays. Separately, 242 real, unedited management decisions power a “guess the model” quiz on Firmulate’s site.
From watching to a company-specific pilot
For a business considering AI in customer service, sales or operations, a company-wide rollout is not the only way to learn what an agent might do under pressure. Firmulate says an enterprise pilot can run the wargame against a read-only export of its own business. The exercise is designed to surface crisis responses, decision-making and weak points in a company’s playbooks, while nothing writes back to real systems.
That creates a practical next step for retailers and other businesses curious about AI agents: bring company context into a controlled exercise, then inspect how the models handle the scenarios that matter to you. The point is to see decisions in context before entrusting agents with day-to-day work.
Try the wargame with your own business
Firmulate’s live experiment shows why recognizing a problem does not guarantee that an AI will act on its own analysis. Enterprises can explore a pilot using a read-only export of their business. Learn more at firmulate.com/pilot.html or 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.
