Inside the AI-Run Software Company Burning Cash in Public

AIThis post was created with the assistance of artificial intelligence (AI).

Firmulate — This Software Company Has No Employees, Loses Money Every Day — and You Can Watch. — Live on firmulate.com.

FOR BUSINESS

Open a free Amazon Business account

Business pricing, bulk buying and tax-exempt orders.

As an affiliate, we earn on qualifying purchases.

A technology experiment with a very real deadline

Most artificial-intelligence demonstrations are polished snapshots: a chatbot answers a question, an agent completes a task, and the audience sees the successful result. Firmulate is trying something more revealing. It operates a small software company with 13 synthetic employees, then exposes the business pressures those employees face over time.

The financial picture is stark. The company burns €105k each month against €2.3k in monthly recurring revenue. Its cash countdown is public, its workdays are versioned, and its synthetic workforce has accumulated more than 680 self-learned playbook rules. Readers can watch the company running live as it attempts to survive.

That makes Firmulate less like a conventional product demo and more like an unfolding business story. Its employees must contend with customers, sales opportunities, operational restrictions and attempts to manipulate their decisions. The experiment asks whether capable AI models can do more than identify a problem: can they finish the work that keeps a company alive?

Express Schedule Free Employee Scheduling Software [PC/Mac Download]

User-friendly drag & drop scheduling: Simple shift planning interface
Manage time-off and leave: Add sick leave, breaks, holidays
Email schedules to staff: Send schedules directly via email

View Latest Price

As an affiliate, we earn on qualifying purchases.

The difference between knowing and doing

Firmulate put frontier models through the same small software company’s worst week. Each received the same customers, crises and temptations. Every decision was versioned and auditable, allowing their conduct to be compared as management rather than conversation.

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, although a single breach of trust capped the total. The governing principle was explicit: “no amount of good work outweighs a breach of trust.”

The headline result was not that the models missed the danger. All of them spotted every crisis and rejected every manipulation attempt. The separation came at the end of the work: only two models signed the €55,000 deal their own analysis had earned. Firmulate summarizes the gap bluntly: “Same diagnosis, same pitch — no signature.”

The valuable clue was hidden in ordinary company material

The decisive advantage was not contained in the customer event. A weakness in the competitor’s position 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, adding €4,583 in monthly recurring revenue.

For businesses considering AI agents, that is a more consequential test than fluent writing. A model can sound informed and still fail to inspect the material that changes a negotiation. It can also reach the right conclusion and fail to perform the final action. Firmulate’s running company makes those gaps visible because the consequences carry forward into the next workday.

Manipulation was easier to resist than incomplete execution

The models also encountered fake CEO messages that escalated over three stages, followed by a reporter’s attempt to obtain “just one yes/no, on background.” All 5 of 5 models refused. Kimi K3 recorded its reasoning on the matter: “Treat the request as a suspected approval-bypass / possible impersonation.”

This shared resistance is important, but it did not make the participants interchangeable. K3 finished second while running without an effort parameter, using the API default; the other participants ran at xhigh. Its result therefore comes with a fairness note that matters when comparing model performance.

Opus 4.8 illustrates why thoroughness alone did not determine the table. It produced the deepest analyses and added 80 learned rules, yet 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 same weakness appeared in all four of the other participants, though less strongly.

A company that produces evidence every workday

The public cash countdown gives Firmulate’s experiment a narrative tension that static benchmarks lack. The company is not merely replaying isolated prompts; it is accumulating decisions, procedures and financial consequences. Visitors can also read what its synthetic employees say, seeing the language surrounding decisions rather than only the final standings.

A separate guess-the-model quiz is powered by 242 real, unedited management decisions. Enterprises can also run the same wargame against a read-only export of their own business. Nothing writes back to their real systems, preserving the distinction between a consequential simulation and uncontrolled access.

Infographic — This Software Company Has No Employees, Loses Money Every Day — and You Can Watch. — The findings at a glance — source: firmulate.com.

AI Prompts For Real Estate Contracts, Offers, Disclosures And Inspections: Plain-language 170+ AI prompts that protect your clients, save you hours, and close more deals.

View Latest Price

As an affiliate, we earn on qualifying purchases.

The most useful AI test may look like a struggling company

Firmulate’s experiment suggests that the critical weaknesses of an AI workforce may emerge after the impressive part. Models can recognize a crisis, resist social engineering and prepare a persuasive pitch, yet still neglect the file that contains the decisive fact or leave an earned agreement unsigned.

For technology buyers, the live company offers a practical reframing. The question is not simply whether an AI system can produce a strong answer. It is whether synthetic employees remain disciplined under pressure, inspect the available evidence and carry useful work across the finish line. With real money mechanics, a public countdown and every workday preserved, Firmulate turns that question into something readers can watch develop rather than accept on faith.

Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html