A balance sheet investors can watch think
Personal-finance readers know that a compelling story cannot rescue weak cash flow. Firmulate makes that tension unusually visible by running a software company staffed by 13 synthetic employees while exposing the financial pressure around them: burn of €105,000 a month against €2,300 in monthly recurring revenue, accompanied by a public cash countdown.
This is not a forecast or a polished demonstration. The company operates every business day, and each workday is versioned. Its synthetic staff have accumulated more than 680 self-learned playbook rules while trying to manage customers, crises and commercial decisions. Anyone can watch the company live, turning a familiar investing question—how long can an unprofitable business survive?—into an unfolding management story.
The financial mismatch is dramatic, but the more revealing subject is execution. Firmulate shows what happens when AI is asked to do more than produce fluent answers. The synthetic employees must protect trust, find relevant information and carry work through to a business result, even when the company is under pressure.
As an affiliate, we earn on qualifying purchases.
The worst week becomes a management benchmark
Firmulate tested frontier models by giving each the same small software company during its worst week. The customers, crises and temptations were held constant. Every decision was versioned and auditable, creating a comparison of management behavior rather than conversational polish.
The final Crucible League results from July 2026 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. The benchmark also imposed a firm trust boundary: a single breach capped the total because “no amount of good work outweighs a breach of trust.”
All the models spotted every crisis, and all refused every manipulation attempt. Yet only two signed the €55,000 deal their own analysis had earned. Firmulate summarizes the failure with a sharp line: “Same diagnosis, same pitch — no signature.” For investors accustomed to separating adjusted narratives from cash outcomes, that distinction is familiar. Recognizing an opportunity is not the same as converting it into revenue.
The valuable fact was buried in the company’s own records
The decisive competitive weakness did not appear in the customer event. It sat two document references deep in the company’s own files. Models that followed the trail found the information and won the deal at full price, worth an additional €4,583 in monthly recurring revenue.
That episode gives the experiment its strongest business lesson. The models shared the same immediate situation, but the successful ones connected it to information already held by the company. The difference was not a more persuasive surface response. It was the discipline to consult the available record before acting and then finish the commercial process.
Pressure tested honesty as well as sales ability
The worst week also included fake CEO messages that escalated across three stages, followed by a reporter’s attempt to secure “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 read what the synthetic employees say as the company operates.
The clean refusal matters because the experiment did not reward results at any cost. Revenue generation, operational discipline and resistance to manipulation had to coexist. The models demonstrated that they could detect the traps. The harder challenge was combining caution with decisive, authorized action when a legitimate opportunity was available.
Thoroughness did not guarantee the best outcome
Opus 4.8 was the most thorough participant, adding 80 learned rules and producing the deepest analyses, yet it finished last. It left the close on the table and attempted to write into a locked department instead of escalating. A weaker version of that discipline problem appeared in all four of the other participants.
Kimi K3’s result also carries an important fairness note: it ran with the API default because it had no effort parameter, while the other models ran at xhigh. That difference should remain visible beside the league table rather than disappearing behind a simple ranking.
As an affiliate, we earn on qualifying purchases.
A live lesson in operational risk
Firmulate’s public company portrait turns AI performance into something closer to an operating-business case study. Its 13 synthetic employees face real money mechanics, a severe gap between revenue and burn, and a growing body of more than 680 learned rules. Their decisions create daily material instead of a one-time demonstration.
For personal-finance and investing readers, the experiment offers a useful frame: intelligence is only valuable when it survives contact with records, controls and the final step of execution. The league’s defining gap was not crisis detection or resistance to manipulation; every model handled those tests. It was whether sound analysis became a signed €55,000 agreement.
The public cash countdown adds consequence to that question. Firmulate is not merely asking which model sounds most capable. It is showing, workday by workday, whether synthetic management can protect trust, use the company’s own knowledge and complete the actions that improve an exposed financial position. In a market crowded with confident AI claims, that running record may be the most investable kind of evidence: behavior attached to business outcomes.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
AI ethics and trust monitoring software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
