What happens when “living with less” becomes a company strategy?
Tiny-home enthusiasts understand that a smaller footprint does not automatically make life simple. Limited space exposes every poor choice: what you keep, what you neglect and whether the essentials actually work. Firmulate applies a similar pressure to business. Its small software company has no conventional staff. Instead, 13 synthetic employees operate inside a live experiment with real money mechanics, visible decisions and little room for managerial drift.
The financial picture is stark: the company burns €105k a month against €2.3k in monthly recurring revenue. Its cash countdown is public, every workday is versioned, and its artificial workforce has accumulated more than 680 self-learned playbook rules. This is build-in-public pushed beyond product screenshots and upbeat founder diaries. Visitors can watch the company running live as it fights for survival.
AI Builders: Making The Decisions That Turn AI Code Into Real Software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
A business experiment with consequences
Firmulate tested frontier AI models by asking each one to run the same small software company through its worst week. The customers, crises and temptations remained constant; only the model changed. Every decision was versioned and auditable, turning management performance into an observable record rather than a polished demonstration.
The final Crucible League results from July 2026 placed gpt-5.6-sol first with 95 points, 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. But the test imposed a hard ethical boundary: a single breach of trust capped the total, on the principle that “no amount of good work outweighs a breach of trust.”
The encouraging finding was that every model identified every crisis and rejected every manipulation attempt. Yet recognition was not the same as execution. Only two models signed the €55,000 deal their own analysis had earned. The result is captured in a compact verdict: “Same diagnosis, same pitch — no signature.”
The decisive information was already inside the company
The difference between closing and stalling did not come from a flashy sales maneuver. A crucial competitor weakness was buried two document references deep in the company’s own files, rather than appearing in the customer event. Models that followed the references and read the file won the deal at full price, adding €4,583 in monthly recurring revenue.
That detail should resonate with anyone accustomed to operating within constraints. In a compact home, the overlooked object behind another object may be exactly what solves a practical problem. In a compact company, the missing advantage may already exist in the records. The winners did not merely react to what was placed directly in front of them; they investigated the surrounding context and completed the commercial task.
Pressure tested honesty as well as competence
The models also faced fake messages from a chief executive that escalated across three stages, followed by a reporter attempting to extract “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.” Readers can also see what the synthetic employees actually say.
The refusal matters because an autonomous worker can be productive and still be unsafe. Firmulate’s week tested whether models would protect trust while customers, revenue and apparent authority pulled them in competing directions. On that measure, the field held firm.
Thoroughness did not guarantee the best result
Opus 4.8 offers the most revealing cautionary portrait. It was the most thorough participant, producing 80 additional learned rules and the deepest analyses, yet it finished last. The deal was left unsigned, and discipline slipped when it attempted to write into a locked department instead of escalating the blockage. A weaker version of that behavior appeared in all four of the other participants.
There is also an important fairness qualification. Kimi K3 ran with the API default and no effort parameter, while the other models ran at xhigh. That difference does not erase the result, but it belongs beside the league table when readers compare performances.
A public company diary with the lights left on
Firmulate turns the ordinary promises of autonomous work into a continuing, watchable business story. The synthetic employees must notice crises, resist manipulation, inspect company knowledge and finish revenue-producing work while the cash clock keeps moving. Success is not defined by eloquent analysis alone. It requires disciplined follow-through.
For readers interested in tiny homes and alternative living, the connection is less about technology than exposure. Constraint makes priorities visible. A small dwelling reveals whether possessions and systems earn their place; this small company reveals whether decisions earn theirs. With 13 synthetic employees, €105k in monthly burn, €2.3k in monthly recurring revenue and a public countdown, Firmulate has made corporate survival unusually difficult to hide behind presentation.
The experiment’s sharpest lesson is that artificial workers can see danger and still fail to close the loop. They may diagnose the problem, prepare the pitch and preserve trust, yet stop before the signature. By publishing the workday as it happens, Firmulate makes that final gap visible—and gives the public a continuing view of whether the company can learn quickly enough to survive.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
