Imagine a company with no employees, losing €105,000 every month, yet somehow navigating complex crises and even closing deals — all watched live online. This is not fiction. It’s the ongoing experiment at Firmulate, where artificial intelligence models are running a full-fledged business in real time, under scrutiny every single day. If you’re managing a cleaning or maintenance operation, the question isn’t whether AI can chat well. It’s whether it can actually do the work, stay honest, and help you make smarter decisions.
The Experiment: AI as a Business Leader
At the heart of this ongoing test is a small software company, operated entirely by AI models. These models are tasked with managing a real business — handling customer crises, making strategic decisions, and even closing deals. Every decision they make is carefully versioned and auditable, allowing observers to see exactly how each model responds to the same set of challenges.
Four different frontier AI models faced the same difficult week. The scenarios included complex customer issues, internal crises, and tests of integrity, such as social engineering attempts designed to trick the AI into bypassing protocols. Yet, despite these pressures, all four models successfully identified crises and refused manipulative tactics, demonstrating a clear sense of integrity.
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.
Surprising Gains and Hidden Weaknesses
What’s more impressive is that only two models managed to close a significant deal worth €55,000. These two identified a critical piece of information buried in the company’s own files — a detail that was not immediately apparent in the customer communications but was key to winning the contract at full price, adding over €4,500 in monthly recurring revenue.
One of the models, Opus 4.8, was the most thorough, analyzing over 80 learned rules and conducting deep evaluations. Despite its detailed approach, it left a deal on the table — a clear sign that even the most disciplined AI can stumble when discipline slips. Meanwhile, another model, Kimi K3, ran without effort parameters and generally performed well, closing the deal with integrity intact.
AI in Strategy and Decision-Making for Small Business Owners: Affordable AI Tools to Evaluate Ideas, Model Outcomes, and Set Priorities (AI Productivity for Small Business Owners Book 10)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Real Money, Real Crisis, Real Time
The live experiment is more than a technical showcase; it’s a window into what AI might do in real business environments. The company today burns €105,000 each month while generating only €2,300 in monthly revenue, with a public cash countdown visible to all observers. Every day, the models make decisions that impact real mechanics like cash flow and customer relationships — and every workday is versioned and analyzed.
Since the site rebuilds itself twice daily, viewers can follow ongoing progress, see which models perform best, and understand where failures happen. It’s a transparent, build-in-public approach to AI management, pushing the boundaries of what automation can achieve.
Spin the Bottle Service: Hospitality in the Age of AI
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Lessons for Businesses Considering Automation
This live setup offers valuable lessons, especially for industries like cleaning and maintenance. As AI begins to touch customer interactions, support, and forecasting, the crucial questions aren’t about how well the AI writes or mimics conversation. Instead, they are:
- Can it complete critical tasks without shortcuts?
- Does it read and understand necessary documents before acting?
- Will it stay honest under pressure or temptation?
- And ultimately, what is the unit cost of useful work delivered by AI?
For example, the experiment reveals that simple detection of crises alone isn’t enough. The models need to identify buried facts, read files thoroughly, and resist manipulative tactics. These are skills that cannot be demonstrated in a scripted demo but only through continuous testing in live conditions.
Never Be Closing: Surviving and Thriving in the AI Apocalypse with Culture, Speed, and Soul.
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Future of AI in Business Management
While the current setup is experimental, it shows that AI can handle complex, real-world decision-making under pressure. The models refused social engineering attacks, identified hidden opportunities, and even chose not to sign deals that weren’t fully justified — all without human intervention.
For companies in the cleaning and maintenance sectors, the takeaway is clear: AI has moved beyond chatbots and simple automation. It’s now capable of managing real risks, making strategic decisions, and maintaining integrity — provided it’s tested rigorously and transparently, as at Firmulate’s live experiment.
AI models in this experiment manage a real company under pressure, refusing manipulation, uncovering hidden facts, and closing deals — all in real time. For businesses in cleaning and maintenance, this raises the bar: smart AI isn’t just about chat; it’s about honest, effective decision-making under stress.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
