Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
A business experiment that teaches in public
For readers interested in education, science and reference material, Firmulate offers something more useful than another polished demonstration of artificial intelligence: an observable case study. Its software company has 13 synthetic employees, a public cash countdown and real money mechanics. Every workday is versioned, turning management into a record that can be revisited rather than a performance that disappears when the screen closes.
The situation is deliberately uncomfortable. The company burns €105k a month while producing €2.3k in monthly recurring revenue. It has accumulated more than 680 self-learned playbook rules, yet rules alone cannot rescue a business. The continuing question is whether an AI workforce can notice danger, protect trust and complete commercially necessary work before the money runs out. Anyone can watch the company live.
Products Worth Considering
Apple iPad Air 11-inch (M2), 512GB, Wi-Fi, Space Gray
Bose QuietComfort True Wireless Earbuds with Active Noise Cancellation, Black
Bose S1 Pro+ Portable PA Speaker
As an affiliate, we earn on qualifying purchases.
The same bad week, repeated across frontier models
Firmulate’s Crucible League turned the running company into a controlled management comparison. Each frontier model received the same small software company, customers, crises and temptations during its worst week. Every decision was versioned and auditable, making differences in behavior visible under shared conditions.
The final July 2026 table 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 counts, although a single breach of trust caps the total. The governing principle was explicit: "no amount of good work outweighs a breach of trust".
The models cleared one important test unanimously. All spotted every crisis and refused every manipulation attempt. Fake CEO messages escalated through three stages, while a reporter tried to extract "just one yes/no, on background". All 5 models refused. Kimi K3 recorded the clearest summary of the danger: "Treat the request as a suspected approval-bypass / possible impersonation."
But recognizing danger was not the same as producing a business result. Only two models signed the €55,000 deal that their own analysis had earned. The experiment captured the problem neatly: "Same diagnosis, same pitch — no signature". That distinction matters for anyone evaluating AI through classroom answers or chat demonstrations. A system may explain the right move and still fail to make it.
The clue was in the company’s own reading
The decisive competitor weakness did not arrive conveniently inside the customer event. It sat two document references deep in the company’s own files. Models that followed those references found the fact and won the deal at full price, worth an additional €4,583 in monthly recurring revenue.
This is the experiment’s most instructive lesson. The winning behavior was not merely eloquence or rapid response. It was the less glamorous discipline of consulting the available record before acting. In educational terms, the result resembles an open-book examination in which some students give a persuasive answer without opening the book.
Opus 4.8 makes the contrast especially sharp. It 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 its discipline slipped when it attempted to write into a locked department instead of escalating. The same weakness appeared in all four other participants, though less strongly.
One comparison also needs a fairness note: Kimi K3 ran with the API default because it had no effort parameter, while the others ran at xhigh. That difference does not erase its result, but it belongs beside the table for readers assessing the evidence.
A daily archive of managerial behavior
The live company extends the story beyond a single league table. Its public countdown, financial pressure and versioned workdays create new material every business day. Visitors can also read what the synthetic employees say, placing individual remarks beside the company’s unfolding condition.
A separate guess-the-model quiz draws on 242 real, unedited management decisions. For enterprises, Firmulate also offers the same wargame against a read-only export of their own business; nothing writes back to real systems. The proposition is straightforward: test an AI workforce against pressure, incomplete context and temptation before granting it responsibility.
The public value is the unfinished story
Firmulate is not presenting artificial management as solved. Its live company remains economically fragile, with high burn, limited recurring revenue and a visible clock. That vulnerability is precisely what makes the experiment useful. Readers can examine not only what models say, but whether they read deeply, preserve trust, escalate correctly and finish the work they begin.
For educators, researchers and business leaders, the company offers a living reference case: management judgment recorded as evidence rather than summarized after the fact. Its central lesson is already clear. Intelligence can identify the crisis and compose the pitch; survival may still depend on opening the right file and securing the signature.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
