Imagine a business without employees, running in the open, constantly battling crises, and losing money every day—all while the public watches. Welcome to the most extreme experiment in building AI-driven companies in real-time.
The Live Experiment: A Company in Crisis
At the heart of this experiment is a small, simulated software company monitored at firmulate.com/live.html. What makes it extraordinary is that it’s staffed by 13 synthetic employees, guided by real money mechanics and a public cash countdown. Every weekday, its decisions are captured, versioned, and analyzed—making it a transparent window into how AI manages business under pressure.
The AI Models Tested in the Crucible League
Four advanced AI models took on the same challenging week, facing identical crises, customer demands, and temptations to cheat. Their performances were scored based on their ability to diagnose, respond, and ethically navigate the company’s struggles. The scores tell a compelling story:
- GPT-5.6-SOL scored 95 and identified a hidden, crucial document that sealed a €55,000 deal, generating an impressive +€4,583 MRR.
- Kimi K3 scored 93 and also signed the deal, with the cleanest discipline and refusal of manipulation attempts.
- Sonnet 5 scored 88 and signed the deal, though with some process slips.
- Fable 5 scored 77; it stopped short, leaving the deal unexecuted despite identifying the opportunity.
- All models demonstrated an ability to detect crises and refuse manipulative social engineering, like fake CEO messages or media tricks.
As an affiliate, we earn on qualifying purchases.
The Hidden Weakness in the Company’s Data
Interestingly, the decisive advantage for some models came from reading deeper into the company’s own files—specifically, uncovering a buried document reference that was not apparent in the immediate customer interactions. This underscores a vital point: the real challenge isn’t just in responding to visible crises but in thorough information processing.
Beyond the Demos: The Reality of an AI-Driven Business
The live experiment isn’t just a game of scores. It reveals the stark reality of a company burning €105,000 monthly against a meager €2,300 in monthly recurring revenue. The experiment’s transparency—every decision, every process, every learned rule—is accessible and auditable, offering a glimpse into how AI might manage actual businesses in the future.
Discipline Under Pressure
The most thorough participant, modeled by Opus 4.8, followed over 80 learned rules and conducted deep analyses. Despite this, it failed to close the deal—illustrating that even rigorous discipline can slip under stress. The experiment shows that real-world management involves more than rules; it’s about discipline, judgment, and honest decision-making.
Why This Matters for Your Smart Home and Business
As smart home devices and automation become more integrated into daily life, understanding how AI performs under pressure is crucial. Will AI support systems stay honest and effective during a crisis? Will they read all necessary information before acting? These questions are not just theoretical—they’re central to building trustworthy AI systems for homes, appliances, or enterprise operations.
The Future of AI as a Business Partner
This experiment illustrates that AI can detect crises, refuse manipulative attempts, and even close deals when provided with the right information. However, it also highlights the importance of thorough data access and disciplined decision-making. The companies of tomorrow will need AI models that can read deeply, stay honest, and deliver useful work consistently—especially when stakes are high.
To see the live company in action and explore the detailed results, visit firmulate.com/live.html. For a deeper dive into the models’ performances and insights, check out firmulate.com/quotes.html.
This experiment offers a rare, transparent look at how AI manages a real-time business crisis, revealing strengths in crisis detection and honesty, but also exposing vulnerabilities in closing deals under pressure. It’s a crucial glimpse into the future of AI-powered enterprise management.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
