What if your business could be run by AI, in full view, battling real crises — and you could watch it happen live?
Imagine a small software company, operated entirely by artificial intelligence models, facing the same tough week as any startup — customer issues, temptations to cheat, internal errors — all under public scrutiny. Now, picture watching this unfold in real time, with every decision, every mistake, and every victory on display. This is not science fiction. It’s the current experiment by Firmulate, a pioneering platform that is running an AI-driven company in public, testing how well these models can handle the rough-and-tumble of real business management.
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 Company: A Digital Reality Show
At the heart of Firmulate’s experiment is a simulated company with 13 synthetic employees, each driven by AI models designed to emulate decision-making processes. The company is juggling real money mechanics: burning through €105,000 each month against a modest €2,300 in monthly recurring revenue (MRR). Every workday, this virtual enterprise is versioned, with decisions and rules tracked for transparency, creating a detailed, auditable record of their actions.
Running this company through its worst week tests the integrity and decision-making skills of four different frontier AI models, each with unique strengths and weaknesses. The models are tasked with managing crises, handling customer requests, and resisting attempts at manipulation — all under the glare of public observation. The results? All four models could identify every crisis and refused every manipulation attempt. But only two managed to close a crucial €55,000 deal their own analysis had earned, demonstrating the difference between diagnosis and execution.
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.
Key Findings: AI Can Spot Crisis — But Can It Seal the Deal?
Across the board, the models proved adept at identifying issues—whether a customer complaint or an internal breach. When fake CEO messages escalated, all five models refused to be duped, citing suspicion of impersonation or approval bypass. This highlights a fundamental strength: AI’s capacity to recognize deception and act conservatively in high-stakes situations.
However, a deeper vulnerability emerged in the company’s own documents. The decisive weakness was buried two references deep within internal files—not in the customer-facing reports. When models read these internal notes, they found the key to closing the deal at full price, adding €4,583 MRR. Those that did not delve deep enough missed out on this opportunity, leaving revenue on the table.
AI: THE PERPETUAL INTERN – Its Brilliance and Failures Share the Same Root
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Transparency and Accountability in AI Decision-Making
What sets this experiment apart is its commitment to transparency. Every decision is versioned and auditable, making it possible to reconstruct the entire decision-making process. This level of openness is rare in AI work and crucial for understanding how these models behave under pressure. For instance, Opus 4.8, the most thorough participant, analyzed over 80 rules and conducted in-depth analysis but still left the close unexecuted, demonstrating how discipline and process discipline matter in AI management.
Another interesting aspect is fairness. Kimi K3 was run with default API settings, while the others used high-effort parameters, which could influence performance. Despite this, K3 managed to close the deal, showing promising discipline and reliability.
EXPLAINABLE AI : Techniques that Meet Auditors’ Needs : Building Transparent, Defensible, and Audit-Ready Artificial Intelligence for Modern Enterprises
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Implications for Business and AI Integration
This experiment is more than a technical showcase; it raises pressing questions for any organization contemplating AI adoption in management roles. If AI agents will interface with your CRM, support queues, or financial forecasts, the concern isn’t just whether they write coherently but whether they can finish what they start, read all relevant information—including internal documents—and resist manipulation under pressure.
As the live experiment demonstrates, even the best models can slip up in discipline or execution, leaving potential revenue unrealized. The takeaway? Rigorous testing, transparency, and understanding AI’s decision pathways are vital before deploying such models at scale.
Watch the Future Unfold
Interested in seeing this in action? Visit firmulate.com/live.html to watch the company’s decision-making in real time. Every workday, a new version of the company’s AI management unfolds, providing a unique window into the capabilities and limitations of AI as a business partner.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
