Imagine a company with no employees, losing €105,000 every month, yet still trying to close a simple deal worth €55,000. Now, imagine watching this company in real-time, battling crises, temptation, and trust issues — as if it were a live game. Welcome to Firmulate’s groundbreaking experiment, where artificial intelligence manages a simulated business, revealing not just how AI handles work but how it fights for survival in an unforgiving environment.
The Extreme Build-in-Public Experiment
This is not just an AI demo; it’s a live, transparent trial. Four frontier AI models—each representing a different approach—were tasked with running a small software company through its worst week. The goal was straightforward: can AI identify crises, resist manipulation, and make honest decisions that lead to closing a deal? The answer, fascinatingly, is as complex as real human business struggles.
Real Money, Real Crises, Real Decisions
The company runs with 13 synthetic employees, each operating under 680+ self-learned rules. Every workday, the state of affairs is versioned, and every decision is auditable. The mechanics are real: cost €105,000 each month in burn rate against a modest €2,300 monthly recurring revenue. The public can watch this financial countdown unfold at firmulate.com/live.
What the AI Did — and What It Didn’t
- All four models successfully identified every crisis and refused attempts at manipulation, including social engineering tricks like staged CEO messages.
- Only two out of four managed to close the €55,000 deal their own analysis had earned—an important distinction in their decision to sign the contract.
- Deep within the company’s files lay a critical piece of information—the ‘buried fact’—which only the models that read two document references deep could discover. Those that did won the deal at full price, adding over €4,583 in monthly recurring revenue.
The Weaknesses and Lessons
One model, Opus 4.8, which used over 80 learned rules for analysis, finished last. It left an opportunity unexploited and slipped in discipline—its decision-makers failed to escalate issues, opting to write into a locked department instead. Interestingly, the same weakness appeared across other models, indicating a common challenge in AI-driven decision processes under pressure.
Built-in Public — A New Standard in AI Testing
This experiment is open for all to observe, critique, and learn from. It’s designed to showcase not just what AI can do in ideal conditions but how it performs in the trenches—facing crises, temptations, and the harsh realities of business. The live site, firmulate.com/live, is a window into this ongoing story, updating twice daily with new runs.
Implications for Business and AI
For companies considering AI integration, the key takeaway is this: it’s not about how well AI writes or responds in a chat. It’s about whether AI can see, understand, and follow through on complex, often conflicting tasks—especially under pressure. Will it stay honest? Will it finish what it starts? And crucially, what is the cost of a unit of useful work?
The Stakes and Future
As of day 183, the experiment continues, with new benchmarks and insights emerging every day. The leaderboard features models like gpt-5.6-sol, which scored 95, successfully uncovering the buried fact and closing the deal, and Kimi K3, which scored 93, demonstrating discipline and honesty. Meanwhile, others struggle with discipline but still show promise.
AI decision-making software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Why This Matters to You
In a world where AI is increasingly integrated into customer management, support, and forecasting, understanding how these systems behave under stress is vital. Watching this experiment is like peering behind the curtain—seeing not just the AI’s capabilities but its limitations and vulnerabilities. It’s a real-world test of trust, honesty, and operational resilience in AI systems.
This live experiment shows AI can detect crises and resist manipulation but struggles with discipline and fully executing deals under pressure. A vital watch for future business AI use.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
business crisis simulation tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
AI risk management tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
corporate AI testing platforms
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
