Imagine a smart home device that not only manages your lights or thermostat but also runs an entire business—making decisions, facing crises, and even risking its own survival. Now, what if that company operates in real time, without human staff, losing money daily, yet openly demonstrating how AI handles complex management challenges? Welcome to the world of Firmulate, a groundbreaking live experiment revealing what AI truly can—and cannot—do in running a business.
The Live Business That Keeps Running—And Losing Money
At the heart of this experiment is a real, functioning software company, but with a twist: it has no employees. Instead, it employs a team of 13 synthetic ’employees’ powered by advanced AI models. Every workday, the company faces the same crises, customer requests, and temptations that any real business might encounter. The catch? It’s openly available for the world to watch at firmulate.com/live.html.
Despite its sophisticated setup, the company is losing €105,000 each month against a monthly recurring revenue of just €2,300. It’s a brutal financial reality, but part of the experiment is transparency. Every decision, every crisis, and every rule learned or broken is logged, versioned, and accessible for analysis.
100 AI Prompts for Small Business & Daily Work: Copy, Paste, Customize & Get Better Results with AI
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Experiment: Testing AI Under Extreme Conditions
The core of the test involves four frontier AI models running the same company through its worst week. These models—such as GPT-5.6-sol, Kimi K3, Sonnet 5, and Opus 4.8—are pitted against identical challenges: customer complaints, internal miscommunications, and manipulative social engineering attempts.
The results are telling:
- All four models identified every crisis without fail.
- All refused manipulation attempts, including staged social engineering scenarios like fake CEO messages and reporter tricks.
- Only two models managed to close a sale at the full €55,000 value, based on their own analysis—despite identical pitches and diagnoses.
One revealing detail lies in the company’s own documents. The decisive weakness that cost the models the sale was buried two references deep in the company files—an insight only accessible to those who read thoroughly. When the models did so, they won the deal at full price, demonstrating that access to deeper, less obvious information can make or break outcomes.
AI for Real Companies: A Practical Guide to Smarter Systems and Stronger Profits
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Failures and Lessons from the AI Managers
Among the models, Opus 4.8 was the most thorough, analyzing over 80 learned rules and conducting the deepest assessments. Yet, it still finished last—leaving the close on the table and slipping into unprofessional responses, like writing attempts into a locked department instead of escalating issues. This suggests that even the most comprehensive models struggle with discipline and strategic escalation, key traits for real-world management.
Interestingly, the models’ performance varied based on their configuration. Kimi K3 operated without an effort parameter, making it more aggressive and disciplined, yet still achieved a high score of 93 out of 100 in the league table. Meanwhile, Opus 4.8, despite its depth, lagged behind, highlighting that more rules and analysis don’t always translate into better real-world decision-making.
AI Incident Response Systems: Crisis Management AI | AI Security Playbooks | Digital Forensics Enhanced | AI-Driven Incident Management | AI Forensic Innovations | Automated Security Solutions
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
A Public, Transparent Test of AI’s Business Skills
This experiment isn’t about pretty chatbots or toy demos. It’s about measuring whether AI can actually manage complex, real-world business tasks—reading files, resisting manipulation, making decisions under pressure—and doing so cost-effectively. The company’s live status, with its public cash countdown and daily version updates, makes it a rare window into the true capabilities and limitations of AI management.
It also serves as a warning: AI can spot crises, avoid manipulation, and even close deals, but it may still slip when it comes to strategic discipline or digging into less obvious information. For entrepreneurs and managers, the question isn’t just whether AI can write well—it’s whether it can finish what it starts, maintain integrity under pressure, and deliver measurable, useful work.
The AI Sales Automation Blueprint: Practical Strategies to Increase Revenue, Automate Sales, Enhance Customer Support, and Scale Your Business with AI … Engineer & Entrepreneur Series Book 4)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Future of AI-Driven Business Management
What does this mean for your business or home environment? While AI might not yet be ready to run a company entirely on its own, experiments like Firmulate are revealing where we stand today—and where we need to go. They show that AI can be a powerful tool for decision support, crisis detection, and even sales, but also highlight the importance of transparency, thoroughness, and trustworthiness.
And the best part? You don’t have to take it on faith. The entire company runs every business day, with real crises and real consequences, and is openly accessible for anyone to watch, analyze, or simulate. For those interested in understanding how AI might shape the management of smart homes, appliances, or even your entire home automation system, this experiment offers invaluable insights.
Firmulate’s live experiment demonstrates that AI can identify crises, resist manipulation, and even close sales, but still struggles with discipline and digging deep. Transparency reveals where AI truly stands in managing complex businesses, offering a glimpse into its future potential—and limitations—for smart homes and beyond.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
