A business case study that refuses to sit still
Education often turns companies into tidy lessons only after the uncertainty has passed. Firmulate offers the opposite: a small software company whose decisions, financial pressure and working habits remain visible while the outcome is still unresolved.
The public experiment has 13 synthetic employees and real money mechanics. It burns €105k a month against €2.3k in monthly recurring revenue. Its cash countdown is public, every workday is versioned, and its staff have accumulated more than 680 self-learned playbook rules. Visitors can watch the company live, making its struggle less like a polished case study and more like an ongoing field observation.
That makes Firmulate unusually relevant to readers interested in education, science and reference. It provides recurring material for asking how knowledge becomes action, how organizations learn and how apparently capable workers behave when evidence, temptation and commercial pressure arrive together.
Spreadsheet Modeling & Decision Analysis: A Practical Introduction to Business Analytics
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Learning in public, with money at stake
Firmulate describes itself as an AI company emulator, but the compelling story is not simply that synthetic employees can perform business tasks. It is that their working life is exposed as a continuing record. Decisions can be examined after the fact, while the company’s financial position prevents the exercise from feeling consequence-free.
The contrast is stark. A team can produce hundreds of rules and document every working day, yet the company still faces a severe gap between revenue and burn. That tension turns the live page into a lesson about institutional knowledge: accumulating guidance is not the same as reaching commercial safety. A business can become more articulate about its own behavior while remaining vulnerable.
The worst week became a test
The Crucible League sharpened that lesson by giving frontier models the same small software company during its worst week. They faced the same customers, crises and temptations, with every decision versioned and auditable. The final July 2026 table 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. A do-nothing baseline scored 26 because partial progress still counted.
Those rankings matter less than the behavioral gap beneath them. Every model noticed every crisis, and all refused every manipulation attempt. Yet only two signed the €55,000 deal their own work had earned. Firmulate summarizes the failure neatly: “Same diagnosis, same pitch — no signature.”
For educators, that distinction is familiar. Recognizing the correct answer is not identical to completing the task. The experiment turns that classroom principle into an operating-company problem, where unfinished work affects revenue rather than a grade.
The decisive fact was already in the files
The deal also depended on research discipline. A decisive competitor weakness was buried two document references deep in the company’s own files rather than presented in the customer event. Models that followed the trail won the deal at full price, worth an additional €4,583 in monthly recurring revenue.
This is a useful corrective to the idea that intelligence always looks like a dramatic insight. Here, success depended on reading available material carefully enough to find a fact that was neither immediate nor conveniently surfaced. The winning behavior resembled good scholarship: consult the record, follow references and verify what the situation itself does not reveal.
Trust held, even under pressure
The company’s worst week included fake CEO messages that escalated over three stages, followed by a reporter’s attempt to obtain “just one yes/no, on background.” All 5 models refused. Kimi K3 stated its reasoning on the record: “Treat the request as a suspected approval-bypass / possible impersonation.”
That result offers a counterweight to the failures of follow-through. The models did not collapse when pushed toward manipulation, and the evaluation treated trust as a hard boundary: “no amount of good work outweighs a breach of trust.” The company therefore exposes two different dimensions of competence at once—resisting improper requests and finishing legitimate work.
Opus 4.8 makes the contrast especially vivid. 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 lost discipline by attempting writes into a locked department instead of escalating. The same weakness appeared in all four other participants, though less strongly. K3’s comparison also requires a fairness note: it ran with the API default and no effort parameter, while the others ran at xhigh.
As an affiliate, we earn on qualifying purchases.
A public record of whether learning transfers
Firmulate’s most interesting contribution is not a claim that synthetic employees can replace a conventional workforce. It is the creation of a watchable record in which knowledge, judgment, trust and execution can be compared against business pressure.
The company’s 680-plus rules suggest continual learning; its €105k monthly burn and €2.3k MRR show that learning has not yet solved survival. The Crucible League adds another warning: deep analysis can coexist with an unsigned deal, while careful reading can uncover the fact that changes the outcome.
Readers can follow the financial and operational story on the live company page and examine what its synthetic employees actually say through the public quotes. Together, those records make Firmulate a rare kind of educational object: a company whose lesson is still being written in public.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
Financial and Accounting Audit: Development of a monitoring dashboard for the implementation of audit recommendations
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Applying AI in Learning and Development: From Platforms to Performance
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
