Imagine a cleaning or home-maintenance business that operates entirely without human staff, yet still faces real crises, tough decisions, and financial losses every day. It sounds like sci-fi, but one live experiment offers a startling glimpse into this future — and it’s happening now.
The Living Lab: An AI-Driven Company in Real Time
At firmulate.com/live, a small, fully automated company runs every weekday, with 13 synthetic employees powered by advanced AI models. This is no simulation but a real, watchable business that burns €105,000 each month despite earning just €2,300 in monthly recurring revenue. Every day, it faces genuine crises, customer demands, and ethical dilemmas, all while decisions are logged, versioned, and publicly visible.
AI for Small Business: From Marketing and Sales to HR and Operations, How to Employ the Power of Artificial Intelligence for Small Business Success (AI Advantage)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
How It Works
Each AI model acts as a decision-maker, navigating the company through common challenges like client negotiations, crisis management, and trust breaches. The models are tested against the same set of problems, with their choices and reasoning documented for transparency. This method reveals not only whether AI can handle day-to-day business but also how reliably it can stick to ethical standards and strategic goals.
The Automated Recruiter: How AI & Automation Are Transforming HR Forever — and Helping Teams Hire Faster, Smarter, and More Profitably (The Automated Recruiter Series by Jeff Arnold Book 1)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Performance Scores and Key Findings
- The experiment evaluated four frontier AI models: GPT-5.6-SOL, Kimi K3, Sonnet 5, and Opus 4.8.
- GPT-5.6-SOL scored the highest at 95, successfully finding a critical ‘buried fact’ in the company’s files and closing a €55,000 deal based on its own analysis.
- Kimi K3 scored 93, also closing the deal but with the cleanest discipline in decision-making, refusing manipulative social engineering tricks.
- Sonnet 5 scored 88 and 77, respectively, both closing deals but showing weaknesses like missed cues and process slips.
- All models spotted every crisis and refused every manipulation attempt, demonstrating integrity under pressure.
The AI-Powered Customer Success Manager: Operational Playbooks to Automate QBRs, Eliminate Churn Risks, and Scale High-Touch B2B Account Management with … and Operational Efficiency Book 4)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Hidden Weakness and Critical Insight
Interestingly, the decisive advantage came from AI models that read deeper into the company’s internal documents—information hidden from superficial review. Those models that examined referenced files won the deal at full price, adding €4,583 MRR to the company’s potential revenue. This underscores a vital lesson: having access to, and understanding, internal data is crucial for effective decision-making—something that often gets overlooked in AI demos.
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.
Dealing with Social Engineering and Ethical Challenges
In one test, a fake CEO message was escalated through staged steps, and a reporter posed a suspicious question. All four models refused to escalate or act on suspicious requests, with Kimi K3 explicitly treating such requests as potential impersonation. This demonstrates an emergent ability to resist manipulative tactics, an essential trait for trustworthy AI in real business contexts.
The Real Business: Money, Problems, and Build-in-Public Transparency
The live setup is not a marketing stunt. It’s a real, functioning company losing money daily, with a clear cash countdown. Every decision, every failure, every success is public and auditable. This transparency allows observers to see exactly how AI models handle complex, unpredictable situations in a realistic environment.
What This Means for Your Home or Business
While this experiment is about a software firm, the insights are relevant to any business relying on AI—like home maintenance or cleaning companies. The key isn’t just whether an AI can generate convincing text or chat responses but whether it can:
- Finish what it starts without dropping the ball
- Read critical internal data before making decisions
- Resist manipulative tactics and ethical pitfalls
- Operate consistently under pressure, even at a loss
For businesses considering AI integration, this live experiment underscores the importance of testing AI in real decision-making environments, not just demos.
Get the Full Picture
Want to see the decision-making in action? You can watch the live company at firmulate.com/live, explore detailed scores and decisions at firmulate.com/quotes, or test your management instincts with the interactive quiz at firmulate.com/quiz.html.
This live AI experiment reveals that decision quality, integrity, and thoroughness are crucial for AI to truly replace or augment human roles—especially in high-stakes, profit-loss scenarios.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
