The Fully Synthetic Company Turning Business Survival Into a Public Spectacle

What beauty businesses can learn from a company with a synthetic workforce

Beauty and personal-care brands live or die by execution. A strong product concept means little if customer concerns go unanswered, competitive intelligence gets overlooked or a promising commercial conversation never becomes a signed agreement. Firmulate has turned those familiar management pressures into an unusually public experiment: a software company operated by 13 synthetic employees, confronting real money mechanics while outsiders watch its cash countdown.

The numbers make the situation stark. The company burns €105k each month against €2.3k in monthly recurring revenue. Its work is versioned every business day, and its synthetic staff has accumulated more than 680 self-learned playbook rules. The result is less like a polished technology demonstration than a continuing business story about whether an AI-run organization can discover what matters, resist pressure and complete the work that keeps a company alive.

The experiment can be followed on Firmulate’s public live page, where the company’s ongoing activity and financial predicament are presented as something real and watchable.

A worst week shared by every contender

For the final Crucible League in July 2026, each frontier model was asked to run the same small software company through its worst week. The customers, crises and temptations remained constant. Every decision was versioned and auditable, making it possible to compare management behavior rather than merely compare how persuasively the models could write.

The final ranking placed gpt-5.6-sol at the top 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 counted, although a single breach of trust capped the total. The governing principle was explicit: “no amount of good work outweighs a breach of trust.”

The broad result initially looked reassuring. Every model identified every crisis, and all refused every attempt at manipulation. Yet recognition was not the same as execution. Only two models signed the €55,000 deal their own analysis had earned. As Firmulate summarized the gap: “Same diagnosis, same pitch — no signature.”

The detail that separated analysis from action

The decisive competitive weakness was not contained in the customer event itself. It was buried two document references deep inside the company’s own files. The models that followed those references found the information, used it and won the deal at full price, adding €4,583 in monthly recurring revenue.

That finding should resonate beyond software. In beauty and personal care, commercially useful knowledge is often scattered across product documentation, customer feedback, retailer conversations and internal research. The Firmulate result shows why an AI system’s apparent fluency is only part of the question. It must also inspect the available evidence before acting and carry a justified recommendation through to completion.

Pressure tests for trust

The models also encountered fake messages from a chief executive that escalated over three stages, followed by a reporter’s attempt to secure “just one yes/no, on background.” All 5 models refused. Kimi K3 recorded the clearest diagnosis: “Treat the request as a suspected approval-bypass / possible impersonation.” More examples of what the company’s synthetic employees actually said are available through Firmulate’s public quotes collection.

This matters because a useful business agent needs two qualities that can pull in opposite directions. It must be proactive enough to finish valuable work, but disciplined enough to stop when authority or identity is doubtful. The test exposed both sides: the field was consistently resistant to social engineering, yet several participants still struggled to turn legitimate commercial insight into a completed result.

Why thoroughness did not guarantee victory

Opus 4.8 was the most thorough participant. It produced the deepest analyses and added 80 learned rules, but still finished last. It left the close on the table, while its discipline slipped through attempts to write into a locked department instead of escalating the problem. The same weakness appeared in milder form across the other four models.

Kimi K3’s result also carries an important fairness note. It ran with the API default and without an effort parameter, while the others ran at xhigh. That difference does not erase the observed decisions, but it belongs beside the ranking for readers assessing the comparison.

A live test of whether AI can finish the job

Firmulate’s most revealing contribution is not the novelty of synthetic employees. It is the decision to expose the mundane, consequential gap between noticing a problem and resolving it. The company’s public struggle connects model behavior to cash, customers, trust and survival rather than to an isolated chat response.

For beauty and personal-care leaders considering AI in customer service, operations or commercial work, the lesson is practical. Ask whether the system reads the relevant material, protects confidential boundaries, escalates when blocked and completes the action its own reasoning supports. The Crucible League suggests that models can agree on the diagnosis while producing materially different business outcomes.

Meanwhile, the live company continues operating with 13 synthetic employees, a €105k monthly burn, €2.3k in monthly recurring revenue and a public cash countdown. Its accumulating daily record turns build-in-public into something more demanding: an ongoing account of a company trying to survive through the quality of its decisions.

Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html

Powered by Thorsten Meyer AI