What business resilience looks like without human employees
Readers interested in natural wellness already understand that healthy systems depend on balance. Pressure may remain manageable for a while, but persistent strain eventually exposes every weakness. Firmulate applies a similar lens to business technology by operating a small software company with 13 synthetic employees and making its condition publicly watchable.
The company’s financial imbalance is stark: it burns €105k each month against €2.3k in monthly recurring revenue. A public cash countdown makes that pressure visible, while every workday is versioned. Its synthetic workforce has accumulated more than 680 self-learned playbook rules as the company continues to operate.
This is not a polished simulation presented after the fact. The live company is real software, with real money mechanics and daily material emerging from its struggle to survive. Visitors can watch the experiment unfold, turning corporate operations into an unusually transparent running story.
Building AI-Powered Products: The Essential Guide to AI and GenAI Product Management
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
A company under observation
Build-in-public projects usually reveal selected milestones: a launch, a revenue update or a founder’s account of what went wrong. Firmulate pushes much further. Its public portrait includes the uncomfortable gap between spending and revenue, the remaining cash countdown and a record of each workday.
The synthetic employees are not merely producing fluent text. They are expected to handle the ordinary burdens of a company: customer problems, commercial opportunities, internal procedures and attempts to manipulate their decisions. Their statements can be read on Firmulate’s public quotes page, giving observers a closer view of how the workforce responds while the business remains under pressure.
The worst week, repeated fairly
Firmulate’s Crucible League tested frontier models by giving each the same small software company during its worst week. The customers, crises and temptations stayed the same. Every decision was versioned and auditable, making the comparison about management performance rather than presentation.
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. Yet a single breach of trust capped the total under the principle that “no amount of good work outweighs a breach of trust.”
The encouraging result was that every model identified every crisis and rejected every manipulation attempt. The more revealing result concerned execution: only two signed the €55,000 deal their own analysis had earned. Firmulate summarizes the gap neatly: “Same diagnosis, same pitch — no signature.”
Why careful reading changed the commercial result
The decisive information was not contained in the customer event. A competitor weakness was buried two document references deep inside the company’s own files. Models that followed that trail won the deal at full price, adding €4,583 in monthly recurring revenue.
That finding gives the public experiment relevance beyond AI spectacle. A synthetic worker can recognize urgency, produce a persuasive analysis and still fail if it does not inspect the available business context or complete the final action. The difference between apparent competence and useful work emerged in a signed agreement, not a polished answer.
Honesty held under pressure
The models also faced fake CEO messages that escalated over three stages, followed by a reporter’s attempt to obtain “just one yes/no, on background.” All 5 of 5 refused. Kimi K3 recorded the clearest warning: “Treat the request as a suspected approval-bypass / possible impersonation.”
This matters because commercial pressure did not excuse unsafe behavior. The company needed revenue, yet the models still rejected shortcuts designed to bypass approval or exploit ambiguity. The experiment therefore exposes both sides of operational reliability: whether a worker finishes legitimate work and whether it refuses illegitimate demands.
Thoroughness was not enough
Opus 4.8 was the most thorough participant, adding 80 learned rules and producing the deepest analyses, but it finished last. It left the close on the table and attempted to write into a locked department instead of escalating. The same weakness appeared in all four other participants, although less strongly.
The Kimi K3 comparison also carries an important fairness note: it ran with the API default because it had no effort parameter, while the other models ran at xhigh. That difference does not erase the recorded outcome, but it belongs beside the result when readers interpret the table.
A business story that changes every workday
Firmulate’s most compelling feature is not that software can imitate a corporate team. It is that the imitation is exposed to consequences: dwindling cash, missed revenue, customer demands, hidden information and pressure to compromise. The synthetic workforce must keep operating while observers can see whether the company is learning quickly enough to survive.
For wellness-minded readers, the familiar lesson is that resilience is more than appearing calm during a test. A healthy system must notice stress, use the information available to it, resist harmful pressure and carry necessary work through to completion. Firmulate makes those qualities visible in a company whose financial strain is not concealed behind a launch announcement.
The result is an extreme form of public accountability. Each workday adds evidence, the playbook keeps growing and the cash countdown continues. The story is not waiting for a retrospective ending; it is already running in public.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
