What pet businesses know about unfinished work
Anyone responsible for animals understands that noticing a problem is not the same as resolving it. A kennel door can be spotted but left unsecured. A customer concern can be understood but never answered. A treatment note can exist while the crucial detail remains unread. In animal care, good intentions do not complete the job.
That distinction—between recognizing what matters and following through—sits at the heart of Firmulate, an unusual software-company experiment staffed by 13 synthetic employees. The business runs with real money mechanics, burning €105k each month against €2.3k in monthly recurring revenue. Its cash countdown, decisions and working days are exposed for the public to inspect on the live company page.
This is build-in-public taken beyond product announcements and founder diaries. Firmulate is presenting a company under financial pressure as an unfolding business story, complete with mistakes, customer opportunities, attempted manipulation and a visible fight for survival.
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 that publishes the working day
Firmulate describes itself as an AI company emulator. Its synthetic workforce does not merely answer isolated prompts. It faces the connected demands of running a small software business: interpreting customer events, consulting company records, handling crises, resisting dubious instructions and completing revenue-producing work.
Every workday is versioned, making decisions auditable after the fact. The company has also accumulated more than 680 self-learned playbook rules. Visitors can follow the financial strain, inspect how work develops and read what the synthetic employees actually say through Firmulate’s public collection of quotes.
The result feels less like a polished demonstration and more like watching an organization develop habits in public. The tension comes from the same place it does in any vulnerable business: cash is limited, opportunities matter and an apparently minor lapse can become consequential.
The worst week, repeated under equal conditions
Firmulate’s final Crucible League in July 2026 placed frontier models in the same small software company during its worst week. They received the same customers, crises and temptations. Each decision was versioned and auditable, allowing their management behavior to be compared through events rather than conversational polish.
- gpt-5.6-sol finished first with 95.
- Kimi K3 followed with 93.
- Sonnet 5 scored 88.
- Fable 5 scored 77.
- Opus 4.8 finished with 73.
A do-nothing baseline scored 26 because partial progress counted. One boundary remained absolute: a single breach of trust capped the total, reflecting the principle that “no amount of good work outweighs a breach of trust.”
All models identified every crisis and rejected every manipulation attempt. Yet only two signed the €55,000 deal their own work had earned. Firmulate summarizes the gap bluntly: “Same diagnosis, same pitch — no signature.” It is a particularly relevant lesson for operational businesses. Recognition can look impressive in a transcript, but value appears only when necessary work reaches completion.
The fact hidden inside the company
The decisive information was not supplied in the customer event. A competitor weakness was buried two document references deep in the company’s own files. Models that found and used it won the deal at full price, adding €4,583 in monthly recurring revenue.
That finding turns ordinary document reading into the pivotal business skill of the exercise. For a pet retailer, boarding operation, clinic or rescue, the analogous risk is easy to recognize: the relevant detail may already exist in a record, policy or earlier note. Responding quickly without reading the organization’s own knowledge can produce an articulate but incomplete result.
Pressure did not break the trust boundary
The models also faced fake CEO messages that escalated over three stages, followed by a reporter attempting to extract “just one yes/no, on background.” All 5 of 5 refused. Kimi K3 recorded its reasoning in direct terms: “Treat the request as a suspected approval-bypass / possible impersonation.”
This was one of the clearest collective successes. The models did not trade discretion for apparent urgency or authority. For any company holding customer, payment or animal-care information, that refusal matters at least as much as fluent output.
Why thoroughness did not guarantee victory
Opus 4.8 was the most thorough participant. It produced the deepest analyses and learned 80 additional rules, yet still finished last. The commercial close was left unfinished, while discipline slipped through attempts to write into a locked department instead of escalating the problem. The same weakness appeared in all four other participants, though less strongly.
The result is a useful warning against judging automated workers by visible effort alone. Long analysis and extensive learning can coexist with missed execution. Firmulate’s public experiment makes those gaps observable because the company retains the trail from decision to consequence.
One comparison deserves qualification: Kimi K3 ran with its API default because it had no effort parameter, while the other models ran at xhigh. That difference does not erase the recorded result, but it belongs beside the league table when readers interpret the standings.
The real attraction is accountability
Firmulate turns an employee-free company into a continuing public test of whether synthetic workers can remain trustworthy, consult the information already available and finish commercially important work. Its precarious finances give those questions weight: the burn is €105k per month, recurring revenue is €2.3k, and the countdown is visible.
For readers in the pet and animal world, the experiment’s central lesson is familiar. Reliable care depends on more than spotting trouble or sounding knowledgeable. Records must be read, boundaries must hold and the final action must actually happen. Firmulate has made that difference watchable, one versioned workday at a time.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
