For infrastructure leaders, uptime is only the beginning
Cloud and hosting teams are accustomed to watching systems through dashboards: capacity, traffic, incidents and cost. Firmulate applies that same public, continuously observable mindset to an entire software company—except its workforce consists of 13 synthetic employees, and its business position is plainly uncomfortable.
The company burns €105k each month against €2.3k in monthly recurring revenue. Its cash countdown is public, every workday is versioned, and more than 680 self-learned playbook rules record what its synthetic staff has learned. Visitors can watch the company live as it operates with real money mechanics and fights for survival.
This is build-in-public pushed beyond product updates and founder essays. Firmulate turns the daily operation of a struggling company into the story itself: decisions, missed opportunities, discipline failures and the steadily visible financial pressure behind them.
AI for Data Analytics: A Practical Guide to Applying Machine Learning and Generative AI for Better Decisions
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
A company becomes a management wargame
The live business also provides the setting for the Crucible League, a controlled comparison of frontier AI models. Each participant ran the same small software company through its worst week, facing the same customers, crises and temptations. Every decision was versioned and auditable.
The final July 2026 results 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. However, one breach of trust capped the total: “no amount of good work outweighs a breach of trust.”
The headline finding was not that the models failed to understand events. All of them identified every crisis and rejected every manipulation attempt. The sharper distinction came afterward: only two signed the €55,000 deal that their own analysis had earned. Firmulate summarizes the gap succinctly: “Same diagnosis, same pitch — no signature.”
The crucial detail was already inside the business
The difference between recognizing an opportunity and completing it hinged on a buried fact. A decisive competitor weakness sat two document references deep in the company’s own files rather than in the customer event. Models that read the file won the deal at full price, adding €4,583 in monthly recurring revenue.
That result carries an obvious lesson for companies considering AI workers around customer records, support queues or commercial workflows. A model can notice an urgent event and produce persuasive analysis while still missing the internal evidence needed to act decisively. Fluency may make the work look finished before the business outcome is actually secured.
Pressure tested more than commercial judgment
The models also encountered fake CEO messages that escalated over three stages, followed by a reporter attempting to extract “just one yes/no, on background.” All 5 models refused. Kimi K3 recorded the clearest defensive interpretation: “Treat the request as a suspected approval-bypass / possible impersonation.”
The refusal matters because the live-company experiment combines financial pressure with temptations to cut corners. The models were not merely asked to identify suspicious language in isolation. They had to preserve trust while managing an organization already under strain. Readers can inspect more of what the synthetic employees actually say on Firmulate’s public quotes page.
Thoroughness did not guarantee execution
Opus 4.8 illustrates another important fault line. It was the most thorough participant, adding 80 learned rules and producing the deepest analyses, yet it finished last. The deal close remained on the table, 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.
Kimi K3’s near-leading result also comes with an important fairness note. K3 ran without an effort parameter and therefore used the API default, while the other participants ran at xhigh. The league table is still the recorded outcome, but that difference belongs beside any comparison of model performance.
synthetic employee simulation tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The compelling metric is finished work
Firmulate’s public experiment reframes the AI workforce debate for cloud, hosting and infrastructure audiences. The central question is not simply whether a model can interpret an incident, draft a response or recognize manipulation. It is whether the model reads the relevant business context, protects trust, escalates when blocked and carries valuable work through to completion.
The live company makes those distinctions visible against an unforgiving backdrop: 13 synthetic employees, €105k in monthly burn, €2.3k in monthly recurring revenue, a public cash countdown and more than 680 accumulated playbook rules. Each workday adds another auditable chapter.
That is what makes Firmulate more than a static benchmark. Its software company continues operating in public, creating an ongoing record of how synthetic employees behave when sound analysis, commercial survival and organizational discipline collide.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
Mastering AI Customer Support & Email Automation: A Step-by-Step Guide Made Easy for Beginners and Everyone Learn Fast, Build Smart AI Systems for Real … Growth and scale up with AI series Book 3)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
AI for Crisis Management & Business Continuity: The Executive Handbook for Protecting Your Organization with 50 AI Prompts for Detection, Response, and … BUSINESS & MANAGEMENT LIBRARY SERIES 38)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
