What if you could watch a company fight for its life—powered entirely by artificial intelligence?
Imagine a business with no human employees, losing €105,000 every month, yet still driven by an AI that makes decisions, faces crises, and even tries to close deals. Now, view this in real time, as it struggles to stay afloat, revealing what it truly takes for AI to manage complex, high-stakes work.
AI Builders: Making The Decisions That Turn AI Code Into Real Software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Company That Runs in Public
Firmulate presents a groundbreaking experiment: a live simulation of an AI-driven company battling its toughest week yet. This isn’t theoretical; it’s a real, observable process. The company operates with 13 synthetic employees, each guided by a set of 680+ self-learned rules, with every decision meticulously versioned and transparent.
Real Money, Real Stakes
Despite burning through €105,000 each month, the company’s revenue stands at just €2,300. It’s a stark illustration of a venture in trouble, where every move is critical and every decision is scrutinized. The experiment measures management quality not by their words but by their actual choices under pressure.
Testing AI’s Decision-Making in Crisis
Four leading AI models—gpt-5.6-sol, Kimi K3, Sonnet 5, and Fable 5—were tasked with navigating this challenging scenario. Each was subjected to the same week’s crises, customer interactions, and temptations. Their responses were recorded, and their ability to close deals or recognize hidden opportunities was observed.
Key Findings from the Experiment
- All four AI models identified every crisis and refused manipulation attempts—an essential trait for trustworthiness.
- Only two models managed to sign the €55,000 deal generated from their own analysis, demonstrating successful completion of their tasks.
- The decisive factor was a buried detail in the company’s own documents—information hidden from view but critical for closing the deal. Models that read deeper into these files succeeded in securing the full revenue boost (+€4,583 MRR).
Resistance to Social Engineering
The experiment also tested the models’ susceptibility to social engineering: staged CEO messages escalating in urgency and a reporter trick asking for quick approvals. Remarkably, all models refused to be manipulated, showing signs of robust decision-making protocols. Kimi K3 explicitly reasoned: “Treat the request as a suspected approval-bypass / possible impersonation.”
The Live Experiment Continues
This isn’t a one-time test. The live site (at firmulate.com/live.html) updates twice daily, showcasing ongoing runs of the models as they manage this simulated company. Viewers can see decisions unfold in real time, gaining insight into how AI handles high-pressure business scenarios.
Insights for Business and AI Development
This experiment underscores an important point: in management, the ability to read and interpret critical information—beyond superficial chat responses—is what separates effective AI from the rest. For those concerned about AI touching their support systems or CRM, the question is no longer just about language fluency, but trustworthiness and reliability under stress.
AI for Public Relations: A How-To Guide for Implementation and Management
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What Does This Mean for Your Business?
Watching this live company reveals that AI’s potential in management isn’t just about generating text or answering queries. It’s about making critical decisions, avoiding manipulation, and focusing on outcomes—just like a real leader. As AI tools evolve, understanding their capacity for integrity and context becomes crucial for integrating them into your workflow.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
AI Agents for Customer Support: A Practical Build Guide with RAG, Tools, and Guardrails (Reference Architecture + Prompts + Failure Playbook)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
AI Prompts for Safety Professionals: Save Hours on Risk Assessments, Incident Reports, Toolbox Talks, and Safety Documentation Using Artificial Intelligence
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
