Your files teach an AI how your business works. So does the way you work with AI — your prompts, patterns, and timing.
Every company keeps records: Claims, contracts, repair logs, Notes. They look boring. They are not.
Those files show how your company handles things. Which supplier failed you. Why you wrote off that account.
But the files are only half of it. The other half is how you work with AI: the order you do things in, the checks you run, the calls you make when the output doesn't match expectation. AI learns from all of it.
Two parties gain from your institutional knowledge. The second is the one that is new: the model's maker can enter your market.
One party gains: you. Every prompt improves institutional knowledge, efficiency, and insights.
People picture documents leaving. That is only part of it. Three things teach a model. The last matters most:
Satya Nadella, Chairman and CEO of Microsoft, calls this “intelligence exhaust” — the agent trail left behind from prompting and re-prompting or query correction.
I learned this lesson at eBags.com, years before
AI. As one of the early
companies selling through Amazon Marketplace, we sold about 300,000 packing cubes a
year on Amazon. Amazon could see what customers bought, which colors and sizes became
best sellers, and which prices and discounts worked best. Then AmazonBasics launched packing
cubes that closely copied ours — down to the zipper styles and curved accent
panels — sold them for less, and used our own product pages to cross-sell
theirs. We sent a cease-and-desist. Amazon pulled them, then relaunched a
slightly modified version within months.
The lesson is structural: when a platform can see what works, it can become your competitor. With shared AI, it can see something even more valuable — how your company thinks and works.
The model runs on machines you control, and you keep the learnings.
What your company knows, never leverages a model your rivals already use – not your data, work patterns, or anything your AI built.
Assume the vendor could one day sell what you sell. Ask whether you would still hand it your playbook.
Asking a vendor not to train on your data is the right move, but that clause is narrower than most buyers assume. See the companion brief, “Your Data Doesn't Train Our Model.” Technically.
eBags account based on Jon Nordmark's firsthand experience and memories as founder and CEO of eBags.
Read: “The Reverse Information Paradox” — Satya Nadella, snscratchpad.com (Jul 12, 2026). Also: “From Static to Dynamic” — Iterate.ai.
A joint educational series on private AI and the AIPod Mini. NetApp — the governed data-control layer. Iterate.ai — the private intelligence layer.
