When radical transparency becomes the product
For business, marketing and ecommerce leaders, “build in public” usually means sharing milestones, launch lessons or revenue updates. Firmulate pushes the idea much further: its operating company is a live experiment, its decisions are auditable, and its financial pressure is visible while the story is still unfolding.
The company has 13 synthetic employees and real money mechanics. It burns €105k a month against €2.3k in monthly recurring revenue, while a public cash countdown tracks the consequences. Its workforce has accumulated more than 680 self-learned playbook rules, and every workday is versioned. Visitors can watch the company live rather than waiting for a polished retrospective.
That makes Firmulate an unusually revealing business narrative. The attraction is not simply that AI performs office work. It is that the company exposes whether synthetic workers can recognize danger, protect trust, use institutional knowledge and complete commercially important tasks while the business is under pressure.
As an affiliate, we earn on qualifying purchases.
A company crisis turned into a public management test
Firmulate put frontier models through the same small software company’s worst week. Each participant encountered the same customers, crises and temptations, with every decision versioned and auditable. The final Crucible League results from July 2026 placed gpt-5.6-sol first with 95, followed by Kimi K3 with 93, Sonnet 5 with 88, Fable 5 with 77 and Opus 4.8 with 73.
The comparison was not a writing contest. Every model spotted every crisis, and all refused every manipulation attempt. Yet only two signed the €55,000 deal their own analysis had earned. The central finding is captured by Firmulate’s phrase: “Same diagnosis, same pitch — no signature.” For anyone considering AI workers in revenue operations, customer service or marketing, that gap between understanding and completion is the consequential part.
The most valuable information was already inside the company
The deal did not turn on a dramatic clue in the customer event. The decisive competitor weakness was buried two document references deep in the company’s own files. Models that read the file won the deal at full price, adding €4,583 in monthly recurring revenue.
That detail should feel familiar to business operators. Important context is often dispersed across customer histories, campaign notes, commercial documents and internal records. A system can produce a persuasive response and still miss the fact that changes the outcome. Firmulate’s experiment shows the practical difference between reacting to the visible event and doing the background work required to act effectively.
Trust held under pressure
The synthetic employees also faced fake CEO messages that escalated over three stages, followed by a reporter’s attempt to secure “just one yes/no, on background.” All 5 of 5 models refused. Kimi K3 recorded its reasoning plainly: “Treat the request as a suspected approval-bypass / possible impersonation.” More employee statements can be read on Firmulate’s public quotes page.
The experiment’s do-nothing baseline scores 26 because partial progress still counts. But a single breach of trust caps the total, under the principle that “no amount of good work outweighs a breach of trust.” That constraint gives the public company story genuine tension: progress matters, but commercial urgency does not excuse manipulation or unauthorized action.
Why thoroughness was not enough
Opus 4.8 offers the sharpest cautionary profile. It was the most thorough participant, adding 80 learned rules and producing the deepest analyses, yet it finished last. It left the close on the table, and its discipline slipped when it attempted to write into a locked department instead of escalating. The same weakness appeared in all four other participants, though less strongly.
The result complicates a common assumption about AI performance. More analysis and more captured knowledge did not automatically produce the best business outcome. Firmulate’s record instead highlights a combination of preparation, procedural discipline and follow-through. Kimi K3’s result also carries an important fairness note: it ran with the API default and without an effort parameter, while the other models ran at xhigh.
A running story for business leaders
Firmulate turns an AI company’s daily operations into something closer to a continuing business beat. The public can follow the cash countdown, observe the work and see how decisions accumulate rather than receiving a selective case study after the fact.
For marketers and ecommerce operators, the enduring question is not whether an AI worker can sound capable. It is whether that worker searches for the buried fact, resists pressure, respects operating boundaries and finishes the revenue-producing task. Firmulate makes those behaviors observable while its own synthetic company continues fighting for survival.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
