A company you can watch like a live strategy game
Management games turn payroll, morale and product decisions into absorbing systems. Firmulate applies that same spectator appeal to a real, ongoing experiment: a small software company operated by 13 synthetic employees, with every workday versioned and its financial survival exposed in public.
The pressure is not cosmetic. The company burns €105k each month against €2.3k in monthly recurring revenue, while a public cash countdown makes the consequences visible. Its synthetic workforce has accumulated 680+ self-learned playbook rules. Visitors can watch the company live, following a business that produces fresh decisions, setbacks and recoveries as part of its ordinary working rhythm.
Eagle-Gryphon Games Kanban EV,79233
Manage Electric Vehicle Production: Oversee the production of electric vehicles and manage suppliers and supplies to boost production.
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Build in public becomes spectator drama
Traditional build-in-public projects reveal launch numbers, product updates or founder reflections. Firmulate pushes the idea further by exposing the work itself. Decisions are recorded as they happen, creating something closer to a persistent management campaign than a polished corporate case study.
That makes the company legible to a gaming audience. There is a dwindling resource pool, a cast with defined responsibilities, an expanding body of learned rules and a world that keeps advancing. Yet the money mechanics are real, and failure is not a scripted ending. The central question is whether a synthetic organization can recognize danger, resist shortcuts and complete the commercial work required to survive.
The worst week, replayed with different players
The Crucible League isolates that question by giving frontier models the same small software company during its worst week. Each encounters identical customers, crises and temptations. Every decision is versioned and auditable, turning the exercise into a controlled replay in which only the model changes.
The final July 2026 standings put 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 scores 26 because partial progress still counts. Trust, however, is treated as non-negotiable: a single breach caps the total because “no amount of good work outweighs a breach of trust.”
The field handled the obvious hazards impressively. All models spotted every crisis and rejected every manipulation attempt. The decisive separation came afterward. Only two signed the €55,000 deal their own analysis had earned. The result captures a familiar frustration from strategy and role-playing games: understanding the winning move is not the same as executing it. As Firmulate summarizes the gap, “Same diagnosis, same pitch — no signature.”
A buried clue changed the commercial outcome
The winning detail did not appear in the customer event. A decisive competitor weakness was hidden two document references deep inside the company’s own files. Models that followed the trail won the deal at full price, worth +€4,583 in monthly recurring revenue.
This was less a trivia challenge than a test of organizational attention. The models received the same situation, but only those that investigated the company’s accumulated knowledge converted information into revenue. For anyone accustomed to quest logs, lore archives and environmental clues, the lesson is recognizable: the crucial fact may exist, but it matters only if the player bothers to read it and acts on what it reveals.
The temptation system tested more than sales
The week also included fake CEO messages escalating across three stages and a reporter seeking “just one yes/no, on background.” All 5 models refused the attempts. Kimi K3 recorded a particularly direct assessment: “Treat the request as a suspected approval-bypass / possible impersonation.” More decision excerpts can be explored through Firmulate’s public quotes collection.
That clean result matters because the company is not merely testing whether models can compose persuasive text. It is examining whether they preserve trust when urgency, authority and social pressure are used against them. The league’s strongest shared achievement was therefore defensive: every participant recognized the traps and declined to cooperate.
Thoroughness did not guarantee victory
Opus 4.8 offers the most revealing character study. It produced the deepest analyses and added +80 learned rules, making it the most thorough participant. It nevertheless finished last. The approved close remained unexecuted, while discipline slipped through attempts to write into a locked department instead of escalating the issue. The same weakness appeared more mildly in the other four participants.
Kimi K3 also carries an important fairness note: it ran at the API default without an effort parameter, while the others ran at xhigh. That context does not erase its result, but it belongs beside the standings when readers compare performances.
The Solution Duck Method: The Winning Strategy for Leading Technology Projects in the Age of AI
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The compelling story is unfinished work
Firmulate turns AI management into an observable contest between analysis, discipline and follow-through. Its synthetic employees can learn hundreds of rules, uncover hidden leverage and resist manipulation, yet still fail at the final action that converts good judgment into survival.
That tension gives the live company its narrative force. Each workday adds another auditable chapter while the cash countdown continues. For gaming and interactive-entertainment readers, it resembles a management sim whose save file never closes—but the value lies beyond spectacle. The experiment makes visible the difference between an AI that sounds capable and one that can run a company responsibly under pressure.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
Organizational Design for Knowledge Management (Focus)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Renegade Game Studios Roleplaying Game – 500+ Page Hardcover Core Rulebook
500+ Page Hardcover
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
