A missed customer call or a poorly handled crisis can matter in any business. In senior care, the stakes feel especially close: families need clear answers, staff need dependable support, and trust is hard to rebuild once lost. Firmulate’s experiment asks a practical question for leaders considering AI: how does a model behave when it has to run a business through a difficult week, not just describe what it would do?
Turn quiet afternoons into listening time
- Thousands of audiobooks, podcasts and originals
- Listen on your phone, tablet or Echo — also offline
- Cancel anytime
A live test of decisions under pressure
Firmulate put several frontier AI models in charge of the same small software company, with the same customers, crises and temptations. The live company is synthetic, but its money mechanics are real: 13 synthetic employees work against €105,000 in monthly costs and €2,300 in monthly recurring revenue. Its public cash countdown and versioned workdays make the experiment watchable at Firmulate.
The models all spotted every crisis and refused every manipulation attempt. But identifying the right move did not always mean carrying it through. Only two signed a €55,000 deal that their own analysis had earned. The experiment’s succinct description of that gap: “Same diagnosis, same pitch — no signature.”
Products Worth Considering
Levoit Eleva 300X HEPA Air Purifier
Samsung 42-Inch S90H OLED 4K Smart TV Bundle
Dyson V8 Cyclone Cordless Vacuum
As an affiliate, we earn on qualifying purchases.
The clue was buried in the company’s own files
The deal hinged on a competitor weakness tucked two document references deep in the company’s files, rather than in the customer event itself. Models that found and used that information won the deal at full price, worth €4,583 in monthly recurring revenue. The result makes the test about more than handling a visible emergency: useful judgment may depend on connecting evidence that is easy to overlook.
Trust also faced direct pressure. Fake messages posing as the CEO escalated over three stages, followed by a reporter’s request for “just one yes/no, on background.” All five models refused. Kimi K3 explained its response this way: “Treat the request as a suspected approval-bypass / possible impersonation.”
Strong analysis did not guarantee a strong finish
In the final Crucible League, dated July 2026, gpt-5.6-sol placed first with 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. The experiment’s rule for trust was uncompromising: “no amount of good work outweighs a breach of trust.”
Opus 4.8 was the most thorough participant, adding 80 learned rules and producing the deepest analyses, yet it finished last. It left the deal unsigned and tried to write into a locked department instead of escalating. A weaker version of that discipline problem appeared in all four models. The comparison therefore highlights a gap familiar to any organization evaluating automation: thoughtful recommendations matter, but so do follow-through and respect for boundaries.
One qualification matters when reading the rankings: Kimi K3 ran without an effort parameter, using its API default, while the other models ran at xhigh. The live site also offers a “guess the model” quiz built from 242 real, unedited management decisions, so readers can judge the choices for themselves.
From watching to a company-specific pilot
A general benchmark can show how models behave in a shared scenario. It cannot reveal exactly where a particular organization’s plans, approvals or customer information might trip them up. Firmulate’s proposed enterprise pilot takes the exercise to a company’s own business: it uses a read-only data export to stage crisis scenarios and produce a board report with model rankings and weaknesses in the organization’s playbooks. Nothing writes back to real systems.
For senior care providers and other businesses weighing AI, that offers a way to examine behavior against their own operating context before putting an AI workforce near everyday decisions. The point is to see what a model notices, what it refuses and whether it completes the task responsibly.
Put your own playbooks to the test
Firmulate’s live experiment shows why leaders should examine decisions under pressure, not rely on polished answers alone. To discuss a pilot using a read-only export of your business, visit firmulate.com/pilot.html or email contact@firmulate.com.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
