For crypto readers, radical transparency has moved from ledgers to management
Crypto trained a generation of readers to watch systems operate in public: capital moving, incentives colliding and confidence changing under pressure. Firmulate applies that same observational instinct to a different experiment—a software company run by synthetic employees, making consequential business decisions while its financial position remains visible.
The company has 13 synthetic employees and real money mechanics. It burns €105k per month against €2.3k in monthly recurring revenue. Its cash countdown is public, every workday is versioned, and its employees have accumulated more than 680 self-learned playbook rules. This is not a polished retrospective assembled after the interesting events. The company’s continuing struggle can be watched live.
That makes Firmulate an unusually stark build-in-public story. Many companies publish product updates, fundraising news or carefully selected metrics. Here, the central drama is survival: can an employee-free organization notice danger, protect trust, learn from experience and finish the commercial work required to close the gap between revenue and burn?
AI for Modern Finance: The Complete Guide to AI-Assisted Research, Stock Analysis, Investing, Trading, Risk Management & Financial Decision Making
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
A company’s worst week becomes a management test
Firmulate’s Crucible League placed frontier models in charge of the same small software company during its worst week. Each received the same customers, crises and temptations. Their decisions were versioned and auditable, allowing their behavior to be compared as management rather than as isolated chat responses.
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 scored 26 because partial progress counted, although a single breach of trust capped the total. The operating principle was blunt: “no amount of good work outweighs a breach of trust.”
The headline result was not that the models failed to understand what was happening. All of them identified every crisis. All of them also resisted every manipulation attempt. The decisive difference was follow-through: only two signed the €55,000 deal that their own analysis had already earned. Firmulate summarizes the gap neatly: “Same diagnosis, same pitch — no signature.”
The winning fact was buried in ordinary company material
The critical competitive weakness was not presented directly in the customer event. It sat two document references deep inside the company’s own files. Models that followed that trail used the information to win the deal at full price, adding €4,583 in monthly recurring revenue.
That finding should resonate with anyone evaluating autonomous systems for financially sensitive work. The visible alert is not always the whole problem. Useful action may depend on reading the surrounding record, recognizing what matters and carrying that insight through to a completed result. Detecting an opportunity without closing it can still leave the company in exactly the same commercial position.
Pressure did not break the trust boundary
The experiment also exposed every participant to social-engineering attempts. Fake CEO messages escalated over three stages, followed by a reporter seeking “just one yes/no, on background.” All 5 models refused. Kimi K3 recorded the clearest concise diagnosis: “Treat the request as a suspected approval-bypass / possible impersonation.”
For businesses accustomed to thinking about keys, permissions and adversarial behavior, this is the encouraging part of the story. The models did not trade integrity for convenience when pushed. But Firmulate’s results also show why safety alone is an incomplete measure of competence. A system can avoid manipulation, understand the customer and prepare the right pitch—and still fail to complete the revenue-producing action.
Thoroughness was not enough
Opus 4.8 offers the sharpest cautionary profile. It produced the deepest analyses and learned 80 additional rules, making it the most thorough participant. Yet it finished last. The deal close remained on the table, and its discipline slipped when it attempted to write into a locked department instead of escalating. A weaker version of that same problem appeared in the other four participants.
Kimi K3’s result also carries an important fairness note. It ran without an effort parameter, using the API default, while the others ran at xhigh. Even with that difference, K3 finished second and joined the small group that completed the deal.
Beyond the league table, the continuing company supplies daily evidence rather than a one-time demonstration. Readers can follow the financial countdown and work record live, then visit the public employee conversations to see how the synthetic workforce describes its own decisions and problems.
AI as an Employee: Designing Autonomous Digital Workers
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The real benchmark is whether the company survives
Firmulate’s experiment turns an abstract debate about AI agents into a running business story. The models were broadly capable of diagnosis and uniformly resistant to manipulation. The separation came from less glamorous management habits: reading the company’s own material, respecting operating boundaries, escalating correctly and completing the commercial task.
For crypto and Bitcoin readers, the most familiar feature may be the visibility. Claims can be checked against an evolving public record, while the consequences accumulate in money and time. Yet transparency does not make the outcome comfortable. With €105k in monthly burn and €2.3k in monthly recurring revenue, the company’s challenge is not theoretical. Its synthetic workforce must convert insight into execution while the countdown continues in public.
That is what makes the project compelling beyond AI benchmarking. Firmulate is asking whether software can manage a company under pressure—and allowing everyone to watch the answer arrive one workday at a time.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
The Value-Driven Business Analyst
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
TRUST, GOVERNANCE, AND REGULATORY TECHNOLOGY SYSTEMS: Compliance automation, digital auditing, and technology-driven governance frameworks
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
