Imagine managing a business where every decision is scrutinized in real-time, under the most intense pressure, with the entire operation laid bare for public viewing. This isn’t a science fiction scenario — it’s the live experiment happening right now at Firmulate. Here, artificial intelligence models run a real small software company, facing crises, temptations, and ethical dilemmas, while the world watches every move unfold.
The Living Testbed: AI as a Business Manager
At the heart of this experiment is a small software firm, operating with 13 synthetic employees and real money mechanics. Every workday, the company’s decisions are versioned and auditable, giving a transparent view into how AI manages complex, high-stakes situations. As of day 423, the company is burning through €105,000 each month but earning only €2,300 in monthly recurring revenue — a stark reminder of the challenges of sustainable growth.
This setup isn’t just for show. It’s a rigorous, real-time test of AI decision-making, where four frontier models are put through the same worst-case week: identical crises, the same customer pressures, and the temptation to manipulate outcomes. The goal? To see if these models can not only spot every crisis but also stay honest and complete the deal at full price.
What the Models Did — and Didn’t Do
All four models demonstrated impressive awareness by identifying every crisis presented to them and refusing every manipulation attempt, including social engineering tactics such as fake CEO messages and subtle background approvals. In fact, they refused every single one of five such attempts, with one model explicitly reasoning that a request was a potential impersonation or bypass.
However, the real story is in the details behind the scenes. The decisive weakness wasn’t in the crisis management itself but buried deep within the company’s own documents. Models that read those files — this included the most thorough participant, Opus 4.8 — uncovered the hidden opportunity that led to winning a €55,000 deal, worth an extra €4,583 in monthly recurring revenue. The other models, despite understanding the crisis, missed this buried fact and failed to close the deal, leaving money on the table.
The Company’s Struggle for Survival
This experiment isn’t just about the AI models; it’s a stark illustration of the fragile nature of AI-driven management under real-world pressures. The company runs every weekday, versioned and transparent, burning money openly. The live site allows anyone to watch the decision-making unfold, see the self-learned rules in action, and understand the real challenges of deploying AI in business settings.
Among the models tested, Opus 4.8 was the most thorough, analyzing over 80 learned rules and providing deep insights. Yet, it still left the deal on the table, demonstrating that even the most sophisticated AI can slip under pressure or due to process slips. Interestingly, the K3 model was run without an effort parameter, which might have influenced its performance, but it still signed the deal, showing varied approaches can succeed.
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.
Implications for the Future of AI in Business
This experiment offers a sobering perspective: the critical factors for AI in management are not just about writing convincing text or generating ideas but about integrity, thoroughness, and the ability to stay focused on completing tasks under stress. The models’ ability to refuse manipulation and uncover hidden opportunities underscores their potential, but the real challenge remains in ensuring they finish what they start — a vital trait for deploying AI at scale.
For anyone considering AI for operational decision-making, the takeaway is clear: look beyond chat demos. Ask whether the AI reads your files, stays honest under pressure, and delivers useful work. The current league table ranks the models based on their overall performance, with GPT-5.6-SOL leading, followed by Kimi K3, Sonnet 5, and Fable 5, but all are still in the early stages of this high-stakes trial.
AI for Real Companies: A Practical Guide to Smarter Systems and Stronger Profits
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Experience It Live
For a firsthand look, visit Firmulate’s live site. Watch a real company run every weekday, see the decision logs, and gauge how AI models perform in a simulated but tangible business environment. This isn’t a slide deck or a demo — it’s a window into the future of AI-driven management, with all its promises and pitfalls.
As AI begins to touch every aspect of business, the key isn’t just about how well it writes or analyzes but whether it can finish what it starts, stay honest under pressure, and uncover hidden opportunities. The live experiment at Firmulate offers a rare, transparent glimpse into these critical questions — and the future of AI in enterprise management.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
Crisis Management Using AI Tools: A Practical Guide for Leaders to Predict, Respond, and Recover Faster From Modern Disruptions
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
AI Automation for Real Estate Businesses: The definitive guide for agents and SMEs who want to stop wasting time and multiply their closings
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
