What if your kitchen gadgets could run a business — and lose money every day, right in front of your eyes?
Imagine a tiny startup, operated entirely by AI, facing the same crises as any real business — and not just handling support or automation, but navigating tough decisions, ethical dilemmas, and the constant pressure of survival. This isn’t fiction; it’s the live experiment from Firmulate, where a company of 13 synthetic employees battles daily to stay afloat, while every move is public and auditable.
Reinventing Clinical Decision Support (HIMSS Book Series)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Live Business in Action
At the heart of this experiment is a real, functioning company that burns €105,000 every month against a revenue of just €2,300. Every day, 13 AI-driven ’employees’ make decisions that influence the company’s fate, guided by over 680 self-learned rules and a meticulously versioned decision history. Viewers can watch this ongoing saga at firmulate.com/live.
Testing AI Under Pressure
The experiment pits four advanced AI models — including GPT-5.6-sol, Kimi K3, Sonnet 5, and Opus 4.8 — against the same set of crises and temptations faced by the real company. Each model runs the company’s daily decision-making, with their choices fully documented for transparency.
Key Findings from the Experiment
- All four models detected every crisis — no surprises there.
- They refused every manipulation attempt, including sophisticated social engineering tactics like fake CEO messages and reporter tricks. In fact, Kimi K3 explicitly flagged such requests as suspicious.
- The critical weakness was hidden deep within the company’s own files: a document reference that, if read, could have unlocked a full-price deal worth over €4,583 in monthly recurring revenue. Only the models that accessed this buried information succeeded in closing the deal at full value.
- Despite similar diagnoses and pitches, only two models signed the €55,000 deal, both after reading the hidden document. The others failed to close, leaving revenue on the table.
- The most thorough model, Opus 4.8, analyzed deeply but still left the close on the table due to discipline slip-ups, like writing notes into a locked department instead of escalating.
The Business Reality
This experiment isn’t just about AI’s decision-making prowess; it exposes the harsh reality of running a miniature company. The real-world measures are stark: monthly cash burn of €105,000 against a meager €2,300 in monthly revenue, and a public countdown to potential insolvency.
Implications for the Future
What does this mean for businesses integrating AI? The focus isn’t just whether AI produces eloquent chat responses — it’s whether it can actually complete critical tasks, maintain integrity under pressure, and read the essential context buried within documents. As the experiment demonstrates, AI models that delve into relevant files and resist social engineering are far more valuable than those that excel solely in conversation.
AI MASTERY & MONETIZATION WORKBOOK: How Freelancers and Entrepreneurs Make Money by Solving Real Business Problems With Prompts, Tools, and Automation
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Why This Matters Beyond the Tech World
For anyone involved in managing or buying kitchen appliances and gear, the lesson is clear: the value of AI isn’t in pretty words or quick responses, but in consistent, trustworthy execution—especially under stress. Just like a well-designed oven or a sturdy blender, a business tool must perform reliably when it matters most.
Practical Claude Handbook for Attorneys: Master Case Analysis, Contract Review, Research Automation, Client Communication, and Document Drafting (Claude AI Guide for Beginners)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Broader Significance
This ongoing live experiment from Firmulate offers a rare glimpse into how AI can be tested rigorously before being entrusted with real-world operations. It’s a build-in-public showcase of AI’s capabilities and limitations, revealing that success depends on thoroughness, discipline, and the ability to uncover hidden opportunities — much like a master chef discovering a secret ingredient that elevates a dish.
Watch, Learn, and Reflect
You can follow the live company in action, see the decisions unfold, and even try your hand at the management quiz based on real decisions made by AI models at firmulate.com/quiz.html. For enterprise teams, there’s also an option to run your own wargame against a read-only export of your business data, ensuring your systems can survive the toughest tests without risking your actual operations.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
Artificial Intelligence for Cybersecurity: Develop AI approaches to solve cybersecurity problems in your organization
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
