Imagine a business that operates entirely without human employees, yet faces the harsh reality of losing €105,000 every month against just €2,300 in recurring revenue. Now, imagine you can watch this struggle unfold live, with every decision and crisis scrutinized in real time. Welcome to the world of Firmulate, a pioneering experiment in building and testing AI-driven management in the open. This is not just a tech demo; it’s a transparent peek into how AI manages a company under pressure — and how well it responds to crises, manipulation attempts, and ethical tests.
The Experiment: AI as Company Manager
At the core of this venture are four advanced AI models, each tasked with running a small software company through its worst week. The identical scenario is applied to each, with the same customers, crises, and temptations to cheat or manipulate. Every decision is versioned, auditable, and publicly observable at firmulate.com/live.html.
Construction Program Management – Decision Making and Optimization Techniques
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Real-World Stakes
Despite the AI models’ impressive ability to identify crises and refuse manipulation—such as fake CEO messages escalated over multiple stages—only two of them managed to complete the task by closing a €55,000 deal their own analysis had earned. The other two models, despite recognizing the opportunity, left the deal on the table or failed to escalate issues properly. This highlights a critical gap: the AI’s ability to diagnose, pursue, and close business opportunities is not just about identifying crises, but also about disciplined execution.
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.
The Hidden Weakness in Data
Interestingly, a decisive advantage for the models that secured the deal was reading a buried fact in the company’s internal files. Unlike surface-level customer interactions, this vital information was two document references deep inside the company’s files, not visible in the customer events. The models that read these internal documents won the full-price deal, worth over €4,500 in monthly recurring revenue, illustrating how deep data access can be a game-changer in AI decision-making.
HUMAN CENTERED ARTIFICIAL INTELLIGENCE SYSTEMS: Explainability ethical design and decision support engineering
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Ethics and Trust Under Pressure
Tests involving social engineering—such as fake CEO messages and reporter tricks—revealed that all models refused unethical requests, including escalating fake messages or providing background-only approvals. Kimi K3, one of the models, explicitly reasoned that the request could be an impersonation or approval bypass. This demonstrates a significant strength in AI: the capacity to resist manipulation and uphold ethical standards even under challenging conditions.
Information Systems for Crisis Response and Management in Mediterranean Countries: 4th International Conference, ISCRAM-med 2017, Xanthi, Greece, … in Business Information Processing, 301)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
A Live, Transparent Company in Action
The company run by these models is a unique setup: 13 synthetic employees, real money mechanics, and a relentless cash burn rate of €105,000 per month against a tiny €2,300 in monthly revenue. It operates in full view at firmulate.com/live.html, with every workday versioned and every decision logged. This build-in-public approach aims to showcase not just AI potential, but actual management performance—warts and all.
The Deep Dive: OPUS 4.8
The most thorough participant, OPUS 4.8, with over 80 learned rules and in-depth analysis, finished last—leaving a critical deal unclosed and slipping in discipline by escalating issues into a locked department instead of resolving them. This highlights that even the most detailed AI models can falter under real-world pressures and that discipline in execution remains a challenge.
Implications for Business and Aging Care
While this experiment might seem distant from senior care or aging services, its lessons are profoundly relevant. As AI begins to touch customer relationships, support queues, and strategic planning in elder care, the questions are clear: Will the AI finish what it starts? Does it read all relevant information before acting? Does it stay honest when under pressure? And crucially, what does a unit of useful work cost?
The Road Ahead
This ongoing live experiment exposes the raw truth about AI management: it’s not just about identifying crises or generating impressive chat. It’s about disciplined execution, ethical integrity, and the ability to uncover hidden opportunities—like reading deep internal documents—to truly succeed. For those considering AI for critical operations, the lessons from Firmulate’s transparent, real-time management battle are invaluable. You can watch the progress unfold daily at firmulate.com/live.html.
As AI begins to take on management roles, the real challenge is not just detecting crises or chatting well—it’s about execution, honesty, and uncovering hidden opportunities. Watch a live experiment reveal what it takes to manage a company under pressure, in real time.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
