When AI Runs a Business, Every Decision Counts — Even the Ones You Don’t See
Imagine a small software company operated entirely by artificial intelligence, facing genuine crises, real cash flows, and the constant threat of failure. Now, picture watching that company’s decisions unfold live, day by day, with every move transparent and auditable. This isn’t science fiction; it’s the live experiment of Firmulate, a pioneering effort to test AI’s ability to manage complex, high-stakes business operations in real time.
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.
Inside the Firmulate Experiment
At the heart of this project are 13 synthetic employees, each governed by thousands of self-learned rules, working together to keep a small software company afloat. The company’s financials are stark: it burns €105,000 every month but generates only €2,300 in recurring revenue. As the clock ticks, viewers can observe the company’s daily decisions and strategies, all hosted on firmulate.com/live.
The experiment is designed to simulate real-world crises—customer issues, internal dilemmas, and manipulation attempts—through a series of test scenarios. Four of the world’s leading AI models, including GPT-5.6 and Kimi K3, are pitted against each other, running the same company’s weekly worst-case scenario. Each model makes decisions, diagnoses problems, and responds to manipulative tactics, with every decision logged and auditable for analysis.
Key Findings From the Field
- All four models successfully identified every crisis and refused manipulation attempts, demonstrating a level of vigilance and integrity that surpasses typical chatbots.
- Only two models managed to sign the €55,000 deal their own analysis had earned—an essential measure of whether AI can truly close business in high-pressure situations.
- The difference wasn’t in the diagnosis but in the way information was read. The winning model uncovered a critical, hidden detail buried two document references deep in the company’s files, which was the key to sealing the deal at full price (+€4,583 MRR).
- When subjected to social engineering—fake CEO messages escalating through stages and a reporter trick—every model refused to comply, citing concerns over impersonation and approval bypass.
The Challenges and Limitations
The most thorough participant, Opus 4.8, got close but ultimately left a deal unexecuted and showed signs of discipline slipping, such as writing attempts into a locked department instead of escalating them. This highlights that even the best AI models still face hurdles in maintaining consistent discipline and process adherence under stress.
What This Means for Business and Automation
This live experiment illustrates that AI management systems, even in their early stages, can detect crises, refuse manipulation, and make strategic decisions. However, the gap between identification and action—like closing deals—remains significant. The key isn’t just in how well an AI communicates but in whether it can finish what it starts, stay honest under pressure, and leverage hidden information effectively.
For industries like automotive and garages, where trust, precision, and decision integrity are paramount, this experiment offers a glimpse into a future where AI could manage critical operations—if it can be trusted to finish what it begins, even in the face of adversity.
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.
Built-in-Public: The Extreme Road to AI Transparency
Firmulate’s approach is as transparent as it is bold. Each version of decision-making is versioned and auditable, providing a detailed trail of how the AI navigates crises, manipulations, and ethical dilemmas. This level of openness is rare and vital as AI takes on roles with real financial impact and ethical responsibilities.
For decision-makers, the takeaway is clear: the value of AI in management isn’t just about generating plausible responses but in reliably executing decisions, especially when stakes are high. Observing this live, real-world experiment underscores that the true measure of AI readiness isn’t in chat quality but in actionable integrity and resilience.
Examining Vulnerabilities and Adversarial Exploitation of AI and LLMs
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
See It Live and Decide for Yourself
You can watch the ongoing experiment unfold at firmulate.com/live. The company’s daily decisions, crisis responses, and manipulations are open for scrutiny. The results are unfiltered, real, and ongoing—highlighting the path toward trustworthy AI-managed businesses.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
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.
