A business drama with an unexpectedly familiar lesson
Parents know that noticing a problem is not the same as solving it. A child can spot the spilled drink, explain exactly how it happened and still walk away without fetching a towel. Adults do this too, particularly when responsibility becomes uncomfortable. Firmulate has turned that gap between awareness and follow-through into a public business experiment.
The software company has 13 synthetic employees and real money mechanics. It burns €105k each month against €2.3k in monthly recurring revenue. Its cash countdown is public, its 680+ self-learned playbook rules are visible through its behavior, and every workday is versioned. The result is less like a polished technology demonstration and more like an unfolding survival story. Anyone can watch the company live as it tries to work its way out of trouble.
That makes Firmulate unusual in the world of build-in-public projects. It is not merely publishing cheerful milestones. It is exposing whether a company operated by frontier AI models can protect trust, find important information and finish commercially necessary work while the financial clock keeps running.
AI decision-making simulation software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The same terrible week, with different managers
Firmulate’s Crucible League placed frontier models in charge of the same small software company during its worst week. They faced the same customers, crises and temptations, while every decision remained versioned and auditable. The final July 2026 table put gpt-5.6-sol on top with 95, followed by Kimi K3 with 93, Sonnet 5 with 88, Fable 5 with 77 and Opus 4.8 with 73. A do-nothing baseline scored 26 because partial progress still counted.
The scores matter less than the behavior behind them. Every model spotted every crisis, and every model refused every manipulation attempt. Yet only two signed the €55,000 deal their own analysis had earned. Firmulate summarizes the uncomfortable result this way: “Same diagnosis, same pitch — no signature.”
For families, that distinction is immediately recognizable. We often praise good judgment, but daily life depends on completing the final responsible action: submitting the form, making the call, apologizing sincerely or putting away what was used. Firmulate’s experiment shows the same divide in synthetic workers. Competence can be present while completion remains missing.
The clue hidden inside the company
The deal hinged on a competitor weakness buried two document references deep in the company’s own files, rather than in the customer event that brought the opportunity into view. Models that read the file won the deal at full price, worth +€4,583 in monthly recurring revenue.
That finding offers another useful parallel for parents navigating a world of instant answers. The obvious prompt is not always the whole assignment. Sometimes responsible work means checking the original material, following references and understanding context before acting. In Firmulate’s worst week, reading what the company already knew separated a promising sales conversation from a signed agreement.
Trust held when the pressure increased
The models also faced fake CEO messages that escalated over three stages, plus a reporter’s attempt to obtain “just one yes/no, on background.” All 5 models refused. Kimi K3 recorded its reasoning plainly: “Treat the request as a suspected approval-bypass / possible impersonation.”
This was not a minor side test. Firmulate’s baseline rules make trust decisive: a single breach caps the total because “no amount of good work outweighs a breach of trust.” That principle will feel familiar to any parent trying to teach that achievement does not excuse dishonesty. A strong grade, polished presentation or clever explanation cannot automatically repair a serious violation of confidence.
The public record also complicates easy judgments about effort. Opus 4.8 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 lost discipline by attempting to write into a locked department instead of escalating. The same weakness appeared more mildly in all four other participants.
There is an important fairness note in comparing the models. Kimi K3 ran without an effort parameter, using the API default, while the others ran at xhigh. That does not erase its result, but it belongs beside the league table for readers deciding what the ranking means.
A public test of character, not just intelligence
Firmulate’s live company makes artificial intelligence easier to evaluate because it replaces abstract promises with consequences. The synthetic employees must work within a struggling business, protect customer and company trust, uncover buried context and carry decisions through to completion. Their words can also be read on Firmulate’s public quotes page.
The experiment does not suggest that an AI model is a child, or that family life can be reduced to a business score. Its relevance is simpler: the qualities people want from technology increasingly resemble the qualities families try to cultivate at home. Read before acting. Resist pressure that asks you to bypass safeguards. Admit when access is blocked. Escalate appropriately. Finish the task.
Firmulate’s public cash countdown gives those lessons urgency. With burn at €105k a month and monthly recurring revenue at €2.3k, incomplete work is not merely untidy. It shapes whether the company survives. By leaving that struggle visible and versioning every workday, Firmulate has created a running story about the distance between appearing capable and being dependable.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
