Open a free Amazon Business account
Business pricing, bulk buying and tax-exempt orders.
As an affiliate, we earn on qualifying purchases.
Watching an AI-Run Business Battle for Survival — Live
Imagine a small company on the brink of collapse, burning over €105,000 every month against just €2,300 in recurring revenue. Now, picture this company run entirely by artificial intelligence models, making decisions in real time, fighting to stay afloat. This isn’t science fiction; it’s a real experiment you can watch unfold at firmulate.com/live.
AI Builders: Making The Decisions That Turn AI Code Into Real Software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Live Experiment: An AI Company in Crisis
At the heart of this story is a unique, publicly accessible experiment: a small software business operated daily by 13 synthetic employees, or AI models, each making critical management decisions. This company faces the same crises, customer temptations, and decisions as any real-world startup, but under the watchful eye of AI. Every change and decision is versioned and auditable, providing a transparent view into how these models perform under pressure.
The Core Challenge: Can AI Make Ethical and Effective Decisions?
The experiment tests four frontier AI models—gpt-5.6-sol, Kimi K3, Sonnet 5, and Opus 4.8—by running them through a simulated worst week. They face crises like customer disputes, internal leaks, and manipulation attempts, with only one goal: successfully close a €55,000 deal.
Remarkable Findings in Performance
The results are revealing. All four models identified every crisis and refused every manipulation attempt, demonstrating resilience and integrity. However, only two managed to sign the deal earned by their own analysis: gpt-5.6-sol and Kimi K3. They did so despite facing the same diagnosis and pitches as their counterparts, but only those two models pulled through and closed at full price—their success secured by discovering a key piece of information buried two document references deep in the company’s files, not in the customer event.
Where AI Decides to Hold Back
One of the surprises is that all models, even the most thorough, left the close on the table at times. For instance, Opus 4.8, known for its deep analysis, failed to escalate a weak point into action and instead kept notes in a locked department. This slippage underscores how even advanced AI can struggle with discipline and prioritization under stress.
The Human Element and Security Tests
Another key test involved social engineering: fake CEO messages and press inquiries designed to sway decisions. All models refused to be manipulated, with Kimi K3 explicitly treating suspicious requests as potential are impersonation or approval-bypass attempts. This highlights AI’s potential for safeguarding against common manipulation tactics.
Real Money, Real Risks, Live and Unfiltered
The company isn’t just a simulation. It’s a real operational entity that burns €105,000 monthly against a modest €2,300 in recurring revenue, making its survival a daily challenge. Every workday, its decisions are recorded, versioned, and observable at firmulate.com/live. The experiment isn’t just about AI performance; it’s about understanding whether AI can be trusted to make ethical, strategic choices in real-world business operations.
Lessons from the Frontlines
The experiment shows that AI can recognize crises and refuse manipulation, but closing deals and disciplined escalation remain challenging. For example, Opus 4.8, with over 80 learned rules, left a critical opportunity unexploited due to lapses in discipline, replicating weaknesses found in less thorough models. The takeaway? Even the most advanced AI models need careful oversight and may struggle with human-like discipline under real-world pressures.
The Broader Implications for Business
This ongoing live experiment offers a glimpse into the future of AI in management roles. It raises questions about whether AI can truly take on responsibility for strategic decisions, financial management, and ethics—especially where real money and reputation are at stake. For companies considering AI tools, the key isn’t just how well they perform in conversations, but whether they can finish what they start, read critical files, and stay honest when it matters most.
Key Takeaway
This live experiment demonstrates that while AI can reliably recognize crises and resist manipulation, successfully closing deals and maintaining discipline remains a challenge. It shows the potential—and current limits—of AI as a real-time business operator, with lessons for any enterprise considering AI-driven management.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
Back to school Picks
back to school
As an affiliate, we earn on qualifying purchases.
