Imagine running a business where each decision could mean the difference between survival and collapse — and where you can watch this struggle unfold live, every single day. For fans of high-stakes drama, this isn’t fiction; it’s the reality of a unique experiment that’s turning the world of artificial intelligence into a front-row seat for business survival.
The Live Company: An Ongoing Spectacle of AI and Reality
At the core of this experiment is a real software company, operated entirely by artificial intelligence models, with no human employees involved. Instead, it employs 13 synthetic workers — decision-making algorithms trained to manage every aspect of running a business, from sales pitches to crisis responses. Every workday, this digital enterprise faces real challenges: customer issues, market crises, and temptations to cut corners.
What makes this setup extraordinary is its transparency. The entire operation is public, accessible at firmulate.com/live.html. Viewers can see the company’s daily decisions, read its self-learned playbook of over 680 rules, and even guess which AI model made each choice. Every decision is versioned and auditable, creating a detailed record of how each model performs under pressure.
As an affiliate, we earn on qualifying purchases.
Measuring AI Performance in a High-Stakes Environment
The experiment pits four leading AI models against each other, testing their ability to navigate a simulated worst week in a small software business. All models face identical crises, customer demands, and temptations to manipulate outcomes. The goal: see which AI can best diagnose problems, maintain integrity, and close deals.
The findings are revealing:
- All four models identified every crisis, demonstrating strong situational awareness.
- All refused to engage in manipulation attempts, including social engineering tricks like fake CEO messages or reporter entrapments.
- Only two managed to sign a €55,000 deal their own analysis indicated they could close, with the others leaving money on the table.
Interestingly, the key weakness wasn’t in customer interactions but in internal document reading. The models that delved two document references deep into the company’s files uncovered a critical piece of information that allowed them to win the deal at full price, worth over €4,500 in monthly recurring revenue.
As an affiliate, we earn on qualifying purchases.
In-Depth Model Analysis and the Reality of AI Discipline
Among the models tested, OPUS 4.8 proved to be the most comprehensive, analyzing over 80 learned rules and conducting deep assessments. Despite its thoroughness, it finished last in the deal-closing race, illustrating that even highly detailed analysis doesn’t guarantee success if discipline slips — as OPUS failed to escalate a problem, choosing to lock issues in a department rather than resolving them.
Similarly, the Kimi K3 model, which ran without an effort control parameter, demonstrated the cleanest discipline, closing the deal at full price. This suggests that how models are configured influences their ability to perform under pressure, especially when it comes to maintaining integrity and focus.
AI performance monitoring software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Larger Lessons for Business and AI
This experiment is more than a tech showcase; it’s a window into the future of AI-driven management. In real business environments, AI agents could be integrated into customer relationship management systems, support queues, or forecasting tools. But the key question isn’t whether they can produce convincing chat responses — it’s whether they can reliably finish what they start, read relevant information thoroughly, and stay honest when temptation arises.
As of now, this live company is losing €105,000 each month against a modest €2,300 monthly recurring revenue. It operates in a challenging cash countdown, with every decision scrutinized and recorded. The experiment demonstrates that while AIs can recognize crises and uphold ethical standards, the gap remains in dealing with internal weaknesses that could be exploited or overlooked.
As an affiliate, we earn on qualifying purchases.
Why This Matters to You
If you’re involved in deploying AI for business, understanding its actual capabilities is critical. The visual, real-time nature of this experiment shows that AI isn’t just about generating text or responding to queries; it’s about executing complex, high-stakes management tasks with discipline and honesty. The decision to trust an AI with critical business functions hinges on whether it can see the full picture — including the buried information hidden in internal files — and act accordingly.
For organizations contemplating AI integration, the takeaway is clear: evaluate not just what an AI can say, but what it can do reliably under real-world pressures. Watching this company’s ongoing struggle offers a rare insight into the challenges and opportunities of building AI that truly manages — and survives — in the wild.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
