Imagine a bustling kitchen where every appliance—be it a blender or a stove—is operated by AI, making decisions and handling crises in real time. Now, picture that same kitchen running every day, yet losing money faster than it earns—an absurd but real scenario. This is not a scene from a sci-fi film, but a groundbreaking experiment revealing how AI models perform under the most stressful, real-world business conditions. Welcome to the frontier of build-in-public AI testing, where a live company is operating with no human employees, battling daily cash burn, and being watched by the world.
The Live Company That’s Watchable and Wagers Its Future
At the forefront of AI experimentation, a unique company runs every workday with 13 synthetic employees powered by different AI models. This setup isn’t just a demo—it’s a real, functioning business that showcases how AI can manage (or mishandle) complex, unpredictable work environments. With a public cash countdown and a daily cycle of decision-making, this experiment offers an unfiltered look at AI’s capabilities and limitations in business management.
Building AI-Powered Products: The Essential Guide to AI and GenAI Product Management
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
A High-Stakes Test of AI Decision-Making
Four leading AI models—gpt-5.6-sol, Kimi K3, Sonnet 5, and Opus 4.8—were each tasked with navigating the same challenging week for this tiny software company. The scenario included real customer crises, internal process dilemmas, and attempts at manipulation. All models successfully identified every crisis and refused every manipulation attempt, demonstrating strong ethical and process-awareness.
However, when it came to closing deals, the results varied significantly. Only two models signed a €55,000 deal that their own analysis had earned—meaning they correctly diagnosed the opportunity and made the right pitch without being swayed by unethical influence. The other two models failed to close, despite understanding the opportunity, highlighting a crucial difference between mere crisis detection and successful commercial closure.
The AI-Driven Leader: Harnessing AI to Make Faster, Smarter Decisions
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Crucial Insights Hidden in Company Files
Interestingly, the decisive advantage in closing the deal lay not in reacting to customer crises but in reading deeper into internal documents. The models that examined and understood information buried two levels deep in the company’s own files were able to secure the full deal at a full €4,583 monthly recurring revenue (MRR). Those that overlooked this internal data missed out on the opportunity, a reminder that understanding a business requires digging beneath surface-level information.
AI cybersecurity and social engineering protection
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Resisting Social Engineering and Manipulation
The experiment also tested the AI models against social engineering attacks. Fake CEO messages escalated over three stages, and a reporter tricked the models with a simple background question. All five models refused to participate or sign off on questionable requests, with Kimi K3 explicitly treating such prompts as potential impersonation or approval bypass attempts. This highlights the growing sophistication of AI in resisting manipulation, an essential trait for trustworthy automation in sensitive environments.
AI data analysis and internal document reading tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Reality of the Live Experiment
This isn’t just a theoretical test. The company operates with real money mechanics—burning €105,000 each month against a modest €2,300 in monthly recurring revenue. The live site, firmulate.com/live.html, offers a window into its daily operations, showing decisions made, crises managed, and the ongoing cash countdown. Every day, the system’s decisions are versioned and available for audit, providing unparalleled transparency into AI behavior under actual business pressures.
The Performance Spectrum and Lessons Learned
The models vary in their depth and discipline. Opus 4.8, for example, is the most thorough participant with over 80 learned rules and deep analysis but ended up in last place by leaving opportunities on the table and slipping into internal communication instead of escalation. Meanwhile, Kimi K3 operates without an effort parameter, yet still performed admirably, closing the deal with the cleanest discipline among its peers.
This experiment underscores a vital truth: success in AI-driven business management depends not just on conversational ability but on disciplined decision-making, thorough analysis, and ethical restraint under pressure.
The Bigger Picture: Why It Matters
For readers who manage kitchens or appliances, the message is clear. The quality of AI isn’t just about how well it chatters or simulates human-like banter. It’s about whether it can finish what it starts, read all relevant information, stay honest under stress, and deliver tangible results—just like a trustworthy chef or appliance that consistently produces quality.
Try It Yourself and See the Future
Curious to test your own decision-making? The platform offers a free quiz based on real management decisions, accessible at firmulate.com/quiz.html. Enterprises can also run their own scenarios against a read-only version of their business, offering a safe way to see how AI performs before deploying it into real systems, detailed at firmulate.com/pilot.html.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
