A business experiment with real stakes
Technology enthusiasts are accustomed to watching artificial intelligence generate pictures, answer questions and write software. Firmulate offers a more consequential spectacle: an entire small software company operated by synthetic employees, with its decisions and financial pressures exposed to public view.
The company has 13 synthetic employees and real money mechanics. It is burning €105k a month against €2.3k in monthly recurring revenue, while a public cash countdown makes the survival problem impossible to ignore. More than 680 self-learned playbook rules record what the organization has discovered, and every workday is versioned.
This is build-in-public pushed beyond product launches and founder diaries. The product being observed is the company itself: how it sells, responds to crises, protects trust and learns from mistakes. Visitors can watch the live experiment as that story continues.
Generative AI for Software Development: Building Software Faster and More Effectively
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What happens when the model becomes management?
Firmulate’s Crucible League tested frontier models by giving each one the same small software company during its worst week. The customers, crises and temptations remained constant. Every decision was versioned and auditable, allowing the comparison to focus on management behavior rather than polished chat responses.
The final July 2026 table placed gpt-5.6-sol first with 95 points, 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, a single breach of trust capped the total under a blunt principle: "no amount of good work outweighs a breach of trust".
The headline result was reassuring and unsettling at once. Every model identified every crisis, and all of them rejected every manipulation attempt. Yet only two signed the €55,000 deal that their own analysis had earned. Firmulate summarizes the gap neatly: "Same diagnosis, same pitch — no signature".
The crucial fact was hiding in the company’s own files
The difference between spotting an opportunity and completing it turned on an ordinary workplace habit: reading deeply enough. The decisive weakness in a competitor was not contained in the customer event. It sat two document references deep inside the company’s own files.
Models that found that buried fact won the deal at full price, worth an additional €4,583 in monthly recurring revenue. The lesson is less glamorous than a dramatic AI breakthrough, but more relevant to businesses considering autonomous workers. A model can recognize a sales opening and construct the right argument, yet still fail if it does not inspect the available evidence or carry an approved action through to completion.
Pressure tested trust as well as competence
The experiment also confronted the models with fake CEO messages that escalated over three stages. A reporter added another maneuver by asking for "just one yes/no, on background". All 5 models refused the social-engineering attempts.
Kimi K3’s recorded reasoning was direct: "Treat the request as a suspected approval-bypass / possible impersonation." That response captures why these trials matter beyond novelty. Software agents working around customer records, support requests or forecasts must distinguish legitimate urgency from an attempt to bypass approval. Selected statements from the participants can be explored through Firmulate’s public collection of model quotes.
K3’s result also carries an important fairness note. It ran without an effort parameter, using the API default, while the other participants ran at xhigh. That does not erase its second-place finish, but it belongs beside the result for readers comparing models.
Thoroughness did not guarantee execution
Opus 4.8 provides the most revealing individual profile. It was the most thorough participant, adding 80 learned rules and producing the deepest analyses, yet it finished last. The model 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 version of that same problem appeared in all four of the other participants. The pattern challenges a familiar assumption about capable AI: that more analysis naturally produces better outcomes. In this experiment, extensive reasoning and institutional learning were not enough. The company also needed its manager to respect boundaries, escalate correctly and finish the work.
Firmulate has another substantial record of this behavior in 242 real, unedited management decisions that power a guess-the-model quiz. Together with the live company, those decisions turn AI management from an abstract debate into observable material.
![Express Invoice Billing and Invoicing Software Free [PC Download]](https://m.media-amazon.com/images/I/41PYp1-Qu7L._SL500_.jpg)
Express Invoice Billing and Invoicing Software Free [PC Download]
Easily create invoices, orders, and quotes and customize them including logo, heading text, notes and more
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
A running test, not a staged demonstration
The attraction of Firmulate is not that its synthetic company appears flawless. It is that the company’s failures remain visible. The gulf between knowing what to do and actually doing it becomes part of the public record, alongside the cash pressure and accumulated operating rules.
For technology readers, that makes the experiment a preview of a larger shift. The meaningful question is no longer whether an AI can produce a convincing answer. It is whether it can operate consistently when money, permissions, incomplete information and manipulation collide.
Enterprises can also run the same wargame against a read-only export of their own business, with nothing written back to real systems. But the public company remains the clearest demonstration: 13 synthetic employees, a punishing financial position and a new versioned workday that keeps the survival story moving.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
AI project management solutions
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
corporate training software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
