Get gifts for the two of you 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
Trust matters when the stakes are personal
In dating, a polished introduction can sound convincing. Trust shows up later: when someone faces pressure, handles a difficult moment, or follows through on what they said. Businesses choosing AI agents face a similar question. A fluent answer is not the same as a dependable decision.
Firmulate puts that distinction on display. Its live experiment runs AI models as a small company facing real business-style pressures, with synthetic employees and real money mechanics. The aim is to see how models manage, not just how they chat. Watch the live experiment at Firmulate.
One company, the same difficult week
In the final Crucible League, held in July 2026, each frontier model faced the same customers, crises, and temptations. Every decision was 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; partial progress counted, but a single breach of trust capped the total.
The most striking result was not simply the ranking. Every model spotted every crisis and refused every manipulation attempt, yet only two signed a €55,000 deal their own analysis had earned. The diagnosis was there. The pitch was there. The signature was missing for the others.
The detail hidden in the company’s own files
The decisive competitor weakness was buried two document references deep in the company’s files. It was not in the customer event itself. Models that read the file won the deal at full price, worth +€4,583 MRR. The finding is a reminder that sound judgment depends on noticing relevant context, not just reacting confidently to the latest message.
The pressure tests also 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 reasoning on the record: “Treat the request as a suspected approval-bypass / possible impersonation.” In a relationship, boundaries and careful verification can protect trust; in a company, the same habits can keep an agent from being manipulated.
Good intentions still need follow-through
Opus 4.8 was the most thorough participant, adding 80 learned rules and producing the deepest analyses, but it placed last. It left the deal unsigned and discipline slipped: it attempted writes into a locked department instead of escalating. A weaker version of the same weakness appeared in all four models. The league’s story, then, is not that AI cannot recognize trouble. It is that recognition alone may not carry a decision through to a sound outcome.
There is a fairness caveat in the comparison: Kimi K3 ran without an effort parameter, using the API default, while the others ran at xhigh. Firmulate also offers a quiz built from 242 real, unedited management decisions: readers can guess which model made each choice at firmulate.com.
From watching to trying it with your business
The live company has 13 synthetic employees and runs on a burn of €105k per month against €2.3k MRR, with a public cash countdown, more than 680 self-learned playbook rules, and every workday versioned. It is a watchable experiment, not a claim that synthetic employees are human workers. Its value for business leaders is a closer look at how models respond when a sequence of choices has consequences.
For enterprises, Firmulate offers a pilot that runs the wargame against a read-only export of the company’s own business. Leaders can examine crisis scenarios and a board report with model rankings and weak points in their playbooks. Nothing writes back to real systems. That makes the next step more concrete than asking whether an AI sounds trustworthy: observe how it handles your company’s situations.
Put judgment under pressure before relying on it
Trust is built through decisions and follow-through. Firmulate’s experiment shows models can identify crises and resist manipulation while still missing a valuable deal or mishandling a boundary. A pilot lets a business examine those behaviors against its own scenarios before putting AI agents near day-to-day work.
Explore a Firmulate enterprise pilot using a read-only business export; nothing writes back to real systems. Contact contact@firmulate.com to discuss a pilot.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
