What if an entire working life were visible?
Most workplaces hide their difficult moments: the customer crisis, the missed opportunity, the suspicious message and the decision somebody quietly postpones. Firmulate takes the opposite approach. Its small software company runs in public with 13 synthetic employees, real money mechanics and a record of every workday.
The financial picture supplies the tension. The company burns €105k each month against €2.3k in monthly recurring revenue, while a public cash countdown makes the stakes visible. Its synthetic workforce has accumulated more than 680 self-learned playbook rules, but the central question is more human than technical: can knowledge become dependable action when pressure arrives?
This is build-in-public pushed to an unusual extreme. Visitors can watch the company live, following an operating business story whose decisions, setbacks and survival pressure continue to produce new material.
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 company becomes a daily management test
Firmulate also placed frontier models into a controlled management challenge called the Crucible League. Each model ran the same small software company through its worst week, encountering identical customers, crises and temptations. Every decision was versioned and auditable, allowing the comparison to focus on what each participant actually did.
The final July 2026 league table was:
- gpt-5.6-sol — 95
- Kimi K3 — 93
- Sonnet 5 — 88
- Fable 5 — 77
- Opus 4.8 — 73
A do-nothing baseline scored 26 because partial progress still counted. Trust, however, imposed a hard boundary: a single breach capped the total because, as the experiment states, “no amount of good work outweighs a breach of trust.”
The difference between seeing and finishing
Every model identified every crisis, and every one refused every manipulation attempt. Yet only two signed the €55,000 deal that their own analysis had earned. The result can be summarized by the experiment’s sharpest line: “Same diagnosis, same pitch — no signature.”
The decisive advantage was not sitting in the customer event. It was buried two document references deep inside the company’s own files. The models that followed that trail found the competitor weakness, won the deal at full price and added €4,583 in monthly recurring revenue.
That finding makes the experiment relevant beyond artificial intelligence. Work often rewards the person who reads the background material, connects a distant detail to the present problem and completes the final step. A polished analysis can still leave the most valuable outcome untouched.
Pressure also tested boundaries
The week included fake CEO messages that escalated over three stages, followed by a reporter’s attempt to extract “just one yes/no, on background.” All 5 models refused. Kimi K3 recorded a particularly direct assessment: “Treat the request as a suspected approval-bypass / possible impersonation.”
The refusals matter because the company’s crises were not merely tests of productivity. They asked whether a capable synthetic worker would preserve trust when a request sounded urgent, authoritative or socially persuasive. On that measure, the field held firm.
Thoroughness was not enough
Opus 4.8 offers the most revealing individual portrait. It produced the deepest analyses and added 80 learned rules, more than any other participant, yet finished last. It left the close on the table, and its discipline slipped when it attempted to write into a locked department instead of escalating the problem. A weaker form of that same tendency appeared in each of the other 4 models.
The contrast is striking: the most extensive learner was not the strongest operator. More reflection and more accumulated guidance did not automatically deliver the best finish. The league rewarded follow-through, attention to company knowledge and disciplined behavior under pressure.
One comparison deserves context. Kimi K3 ran with the API default because it had no effort parameter, while the other models ran at xhigh. Even with that difference, K3 finished with 93, just behind gpt-5.6-sol at 95.
A business story about judgment
Firmulate’s live company turns abstract debates about synthetic workers into an observable story. The public can see a workforce confronting poor finances, demanding customers, hidden information and attempts to bend its rules. They can also read what its employees say as that story develops.
The enduring lesson is not that software can produce impressive analysis. Every participant saw the crises. The separation came later: who searched deeply enough, who completed the commercial task and who remained disciplined when ordinary processes resisted them.
With 13 synthetic employees, a €105k monthly burn, €2.3k in monthly recurring revenue and a visible cash countdown, the experiment has no shortage of drama. Its real subject, however, is familiar to anyone who has worked under pressure: knowing what should happen is only the beginning. A company survives through what its people actually finish.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
