Imagine your favorite kitchen gadget company managed by AI — and it’s struggling to stay afloat.
Now, picture watching that AI company endure crises, make tough decisions, and even face attempts at manipulation — all live on the web. Welcome to the world of Firmulate, where a tiny, employee-less software company is put through its paces in a public, real-time experiment, revealing what AI can really do when tasked with running a business.
The Live Experiment: An AI Company in Action
At the heart of this experiment is a virtual company with 13 synthetic employees, each guided by complex, self-learning rules. Every day, it faces the same set of crises, customer requests, and temptations to cheat — all staged to test the limits of AI decision-making. The company’s financials are public, with a monthly burn rate of €105,000 against just €2,300 in recurring revenue, making every decision critical.
Every workday, the AI models receive new data, and their decisions are recorded and versioned for transparency. This allows observers to see how different models perform under identical conditions. The most recent leader in the experiment, GPT-5.6-sol, scored 95 out of 100, successfully uncovering hidden information in the company’s files that led to a €55,000 deal, significantly boosting revenue by over €4,500 in monthly recurring income.
What This Means for Business and AI
While the experiment may sound like a game, it reveals crucial insights: AI models can identify hidden opportunities, refuse manipulation, and make honest decisions even under pressure. The top-performing models demonstrated they could read between the lines and uncover facts buried deep in company files — a skill that could be invaluable in real-world business scenarios.
But the experiment also highlights the gaps. The most thorough analysis, from Opus 4.8, failed to close a deal because of discipline lapses, illustrating that even advanced AI can stumble without proper constraints. For companies considering AI to manage customer relationships, supply chains, or strategic decisions, the key questions are: Will it finish what it starts? Will it stay honest when tempted? And what is the true cost of useful work?
All of this is happening live at firmulate.com/live, where anyone can watch the journey unfold. This build-in-public approach doesn’t just showcase AI; it tests its limits in the real world of business.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
AI in Property Management: A Practical, Unboring Look at Artificial Intelligence in the Multifamily Industry
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The AI-Driven Leader: Harnessing AI to Make Faster, Smarter Decisions
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Python for AI and Data Analysis: The Hands-On Guide to Data Wrangling, Machine Learning, and Automation with Pandas, NumPy, and Matplotlib (Rheinwerk Computing)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Intelligent Ledger: How Agentic AI, Quantum Computing, Data, and Tokenisation Are Transforming Banking
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
