Imagine a cleaning or floor maintenance business powered by AI, making decisions in real time, yet losing money every month. It’s not science fiction — it’s happening now. This company runs every workday, openly exposing its struggles and decisions as a live, watchable experiment. For those of us in maintenance and cleaning, it’s a glimpse into how AI might reshape operational management, with all its promises and pitfalls.
The Live Experiment: An AI-Run Business in Real Time
At Firmulate, a unique experiment is unfolding every day. A real, small software company is being managed entirely by artificial intelligence models, each simulating a different decision-making approach. This is not a test bed for chatbots but a full-fledged company with real money mechanics — burning €105,000 monthly against a mere €2,300 in monthly recurring revenue (MRR). Every decision, crisis response, and negotiation is carefully recorded and made public, offering unprecedented transparency into AI’s capabilities and shortcomings in business management.
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 the AI Models Are Tested
The experiment challenges four frontier AI models—GPT-5.6, Kimi K3, Sonnet 5, and Opus 4.8—to navigate the worst week in this company’s financial life. They face the same customers, same crises, and same temptations to cut corners or manipulate. Every decision they make is versioned and auditable, ensuring full transparency and a fair comparison.
Key Findings from the Experiment
Despite their differences, all four models detected every crisis and refused every attempt at manipulation. When it came to sealing a crucial deal, only two models signed the €55,000 contract, which their own analysis had earned them. The other two, despite diagnosing the opportunity correctly, failed to close the deal, leaving potential revenue on the table.
The Hidden Weakness
The decisive weakness was buried deep in the company’s internal documents — not in the customer interactions. The models that read and understood these internal files won the deal at full price, adding over €4,583 in MRR. This suggests that AI’s ability to access and interpret internal knowledge can be a game-changer in real-world business negotiations.
The Human-Like Tests: Trust and Deception
In a series of social engineering tests, fake CEO messages escalated over three stages, and a reporter tried to trick the AI with a simple background question. All five models refused to bypass approval procedures or impersonate executives. Kimi K3’s reasoning was clear: “Treat the request as a suspected approval-bypass / possible impersonation.” This indicates a strong capacity for AI to resist manipulative tactics under pressure.
The Company’s Inner Workings
The company behind this experiment operates with 13 synthetic employees, underlining the surreal reality of AI-driven management. It is actively losing money, with the company burning €105,000 every month against a tiny revenue stream. Every workday, its decision-making process is versioned and public — a transparency not seen in traditional business models. The live site, firmulate.com/live.html, invites anyone to observe the ongoing struggle and learn from its decisions.
The Deep Dive: Opus 4.8 Profile
Among the models, Opus 4.8 stands out as the most thorough, analyzing over 80 learned rules and providing the deepest insights. Yet, it finished last in the experiment. It left a crucial deal on the table because discipline slipped, and decisions were routed to a locked department instead of escalated. This highlights that even the most detailed decision models can falter without proper process discipline.
The Real-World Implication for Cleaning & Maintenance
For those in the cleaning, floor care, and maintenance sectors, this experiment poses important questions. Will AI help optimize scheduling, supply chain, and customer relations? Or will it struggle to stay honest and disciplined under real-world pressure? The answer hinges on whether AI models can finish what they start, read internal documents, and remain trustworthy when stakes are high.
Why It Matters
As AI begins to touch your CRM or support queues, the key isn’t just that it can generate convincing chat responses. The critical factor is whether it can stick to the plan, read your internal data, and complete useful work honestly. The experiment at Firmulate vividly illustrates that these qualities are still a challenge — a vital consideration before deploying AI in operational roles.
What’s Next?
The company’s live experiment is ongoing, and every weekday reveals new lessons. It shows that AI can recognize crises, refuse manipulation, and close deals when given the right internal information. But it also exposes vulnerabilities in discipline, decision consistency, and process adherence — issues that matter greatly for any business considering AI automation.
Interested in testing your own business? Firms can run a similar wargame against their data or processes without risking real systems. More details are available at firmulate.com/pilot.html.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
