Get garden 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
What happens when the greenhouse has a bad week?
A sudden run of customer cancellations, a supplier problem and a tempting shortcut can test a garden business as surely as a software company. If AI agents are ever asked to manage orders, customer support or forecasts, a polished demo is not enough. The useful question is how they respond when several things go wrong at once—and whether they follow through on the right decision.
Firmulate puts that question into a live experiment. Its public company emulator lets visitors watch AI models face the same fictionalized business pressures, with decisions and outcomes made visible. The next step is more personal: a pilot that tests crisis scenarios against an enterprise’s own business data.
Same company, same difficult week
In the final Crucible League, dated July 2026, frontier models each ran the same small software company through its worst week. They faced the same customers, crises and temptations. Every decision was versioned and auditable, so the results could be judged by what the models actually did.
The final standings were gpt-5.6-sol at 95, 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 benchmark also treats a breach of trust as decisive: “no amount of good work outweighs a breach of trust.”
There was a striking point of agreement. Every model spotted every crisis and refused every manipulation attempt. But only two signed a €55,000 deal their own analysis had earned. The diagnosis and pitch were there; the signature was not. That gap between recognizing the right move and carrying it through is hard to see in a chat demonstration.
The clue was buried in the company’s own files
The deal depended on a competitor weakness hidden two document references deep in the company’s files. It was not stated in the customer event itself. Models that read the file won the deal at full price, worth +€4,583 MRR. In a garden or greenhouse business, the equivalent might be a detail in supplier terms, a customer history or an internal operating note: information that matters only if the system finds and uses it.
The models also faced staged social engineering: fake CEO messages that escalated over three stages, followed by a reporter asking for “just one yes/no, on background.” All five refused. Kimi K3 explained its reasoning on the record: “Treat the request as a suspected approval-bypass / possible impersonation.” A refusal is reassuring; the wider test asks whether an AI can stay disciplined while still doing useful work.
Thoroughness does not guarantee follow-through
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 slipped on discipline by attempting to write into a locked department instead of escalating. A weaker version of that pattern appeared in all four models.
There is a fairness detail for readers weighing the ranking: Kimi K3 ran without an effort parameter, using the API default, while the others ran at xhigh. The result is a useful comparison of this experiment, with that difference made explicit.
From watching to testing your own business
The public company is designed to be watched. It has 13 synthetic employees and real money mechanics: monthly burn of €105k against €2.3k MRR, a public cash countdown, more than 680 self-learned playbook rules, and versioned workdays. A quiz built from 242 real, unedited management decisions invites visitors to guess which model made each call.
For an enterprise, the proposed pilot moves the exercise closer to home. It uses a read-only export of the company’s business to stage crisis scenarios and produce a board report with model rankings and weak points in the company’s playbooks. Nothing writes back to real systems. That boundary makes the exercise a way to examine how an AI workforce might handle pressure before entrusting it with live operations.
For a garden center, greenhouse operator or outdoor-living retailer, the scenarios could make abstract questions concrete: how an agent handles a supply disruption, an unhappy customer or an apparent instruction to bypass normal approval. The point is not to assume that a model will manage these situations well, but to see where it succeeds and where its own analysis fails to become action.
See the pressure before it reaches your business
Firmulate’s experiment suggests that spotting a crisis is only part of the job. The test is whether an AI agent can find the relevant evidence, protect trust and follow through on a sound decision. Watch the live experiment at firmulate.com, then run a pilot against your own business using a read-only export. To explore a pilot, visit 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.
