Imagine a company with no human employees, losing money every day, yet somehow navigating crises, making decisions, and even closing deals — all under the watchful eye of AI models. This isn’t science fiction; it’s the live experiment run by Firmulate, a pioneering project that exposes how AI can manage a real business in real time, with transparent decision-making and full accountability.
The Experiment: An AI-Driven Company in Its Hardest Week
At the heart of this bold venture is a small software company operated entirely by artificial intelligence models. It’s not a simulation but a live, ongoing experiment accessible at firmulate.com/live.html. Every workday, the company is run through its worst week — facing the same customers, crises, and temptations, with decisions meticulously versioned and auditable, ensuring complete transparency.
The Models Behind the Decision-Making
The experiment pits four frontier AI models against each other, each with distinct profiles and capabilities. These models are rated on their performance using a scoring system — with GPT-5.6-sol leading at 95 points, Kimi K3 at 93, Sonnet 5 at 88, and Fable 5 at 77. The scores reflect each model’s ability to identify crises, resist manipulation, and close deals.
In the final league, the models demonstrated remarkable resilience: all detected every crisis, and all refused manipulative tactics designed to bypass controls. Yet, only two models succeeded in closing the most lucrative deal of €55,000, which their own analysis had earned them. Interestingly, the crucial difference lay in accessing a buried document reference deep within the company’s files — a detail that only the top-performing models read and leveraged to win the contract for an additional €4,583 monthly recurring revenue.
The Reality of a Company Burning Cash
This enterprise spends €105,000 every month in operational costs against a mere €2,300 in monthly recurring revenue, creating a cash countdown visible to all. The company is run with over 680 self-learned rules and every decision is versioned daily, making the entire process auditable and transparent. It operates as a ‘build-in-public’ experiment, exposing both its strengths and weaknesses in real-time.
What Makes This Different? Building Trust and Discipline
The experiment also tested social engineering tactics, such as fake CEO messages escalating through various stages and a reporter trick requiring just a yes/no response on background. Alarmingly for any real-world business, all models refused to be manipulated or impersonated — signaling a promising level of discipline and integrity.
Insights from Deep Analysis
Among the models, Opus 4.8 showed the most thorough analysis, learning over 80 rules and providing the deepest decision dives. Still, it left potential deals unclosed, illustrating that even the best analytical models can slip in discipline when faced with real-world pressures. Meanwhile, Kimi K3, which ran with default settings, achieved the most discipline and closed the deal at full price.
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.
Implications for the Future of Business and AI
This live experiment holds profound implications for industries reliant on AI for decision-making. It showcases that models can be resilient to crises, resistant to manipulation, and capable of complex tasks like closing deals and reading critical internal documents. The question for businesses, especially those dealing with sensitive data and high-stakes operations, is: can AI operate with the same integrity and discipline in your systems?
As AI’s role expands into customer support, sales, and operations, the ability to verify that AI agents can finish what they start — reading files thoroughly, resisting temptations, and maintaining honesty under pressure — will be a critical measure of success.
AI for Data Analytics: A Practical Guide to Applying Machine Learning and Generative AI for Better Decisions
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Why You Should Care
For crypto and Bitcoin enthusiasts, the core lesson transcends the specifics of AI decision-making. It’s about trust, transparency, and accountability in automated systems. Just as blockchain promotes trust through public, immutable records, this experiment demonstrates that AI decision processes can be openly audited and scrutinized. If AI can be trusted to manage a cash-burning enterprise in real time, it could be the foundation for more secure, transparent digital economies in the future.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
AI cybersecurity and manipulation resistance
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
![Claude AI for Beginners Bible: [5 in 1] The Ultimate Guide to Automate Your Work, Save Hours Every Week, and Use AI for Real-World Results](https://m.media-amazon.com/images/I/415+fSJacsL._SL500_.jpg)
Claude AI for Beginners Bible: [5 in 1] The Ultimate Guide to Automate Your Work, Save Hours Every Week, and Use AI for Real-World Results
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
