How AI Read Your Files to Win Business Deals — and Why It Matters for the Future of Work

AIThis post was created with the assistance of artificial intelligence (AI).

Firmulate — We Buried a €55,000 Fact Two Documents Deep. Here's Which AIs Did Their Homework. — Live on firmulate.com.

Imagine an AI that doesn’t just answer questions or browse the web but actually digs into your company’s confidential files to make decisions — and does so accurately enough to close multimillion-euro deals. This isn’t science fiction; it’s the reality unveiled by a groundbreaking experiment where AI models were tested in a simulated business crisis. The result? The difference between winning and losing a €55,000 deal wasn’t just skill — it was whether the AI read beneath the surface, two documents deep, to uncover hidden truths.

In a world where AI tools are increasingly integrated into everyday business operations, the question is no longer whether they can generate convincing text or handle simple tasks. Instead, the critical metric is trust: will these AI agents read and understand your internal files before making decisions? And as a recent live experiment by Firmulate demonstrates, the ability to dig deeper into your company’s own information can be the deciding factor in a competitive bid.

The experiment pitted four state-of-the-art AI models against a simulated small software company facing a tough week of crises, customer interactions, and ethical dilemmas. Each AI was tasked with navigating the same set of challenges, including crises that required not just surface-level responses but a detailed understanding of internal documents buried two references deep in the company’s files.

What did the models do? All four identified every crisis and refused manipulation attempts, such as fake CEO messages and reporter tricks. But only two of them closed the deal and signed a €55,000 contract — the payoff for reading the company’s internal files thoroughly enough to uncover a crucial hidden fact. This buried information in the documents was the key to winning the deal, yet it was invisible in traditional chatbot demos that only skim the surface.

Why is this important? Because in real business, the difference isn’t just in how well an AI can chat but whether it can read your internal data before making a decision. If your AI support or CRM system only reacts to what’s visible on the surface, it might miss the critical details that make or break a deal. Conversely, an AI that reads deeply and understands your files could unlock new levels of trust and efficiency, closing deals at full price and avoiding costly mistakes.

The experiment also tested social engineering scenarios, like staged CEO approval requests and background questions from reporters. All models successfully refused these manipulative attempts, indicating a baseline of ethical behavior and resistance to deception. Yet, the real differentiator was the depth of analysis — the models that went beyond the surface and uncovered the buried fact achieved the full deal, worth an additional €4,583 MRR in the live business simulation.

This live setup isn’t just academic. Firmulate’s platform runs an emulated company in real time, with 13 synthetic employees, real money mechanics burning €105,000 a month, and over 680 self-learned playbook rules. Decision-making is fully versioned and auditable, providing a transparent window into how AI agents operate under pressure. The site, firmulate.com/live, offers a watchable demonstration of this ongoing experiment, showing how AI can perform in a realistic, high-stakes environment.

Among the models tested, the most thorough was Opus 4.8, which analyzed over 80 learned rules and performed the deepest evaluation. Despite this, it still left a deal on the table — a reminder that even the most diligent AI can slip if it doesn’t stay disciplined or escalate properly. Meanwhile, newer models like Kimi K3 managed to close the deal with the cleanest discipline, illustrating that reading depth alone isn’t enough; discipline and process adherence matter as well.

What does this mean for businesses considering AI? It’s not about the shiny demo or natural language fluency. It’s about whether the AI will finish what it starts, read your files before responding, and stay honest under pressure. A unit of useful work isn’t just about generating text — it’s about quality, trustworthiness, and thoroughness. The models that read deeply and act responsibly will be the ones winning the deals and building the trust that sustains long-term success.

To test your own AI’s capabilities, you can try a management decision quiz at firmulate.com/quiz.html. For organizations ready to simulate their own business environment safely and without risk to real systems, a wargame platform is available at firmulate.com/pilot.html. These tools offer a glimpse of how AI can be challenged to perform reliably in the complex, document-rich world of real business.