Last year I filed a patent application for something I’d built: an AI system that answers questions about your own life — your messages, your files, your history — without any of that data ever leaving your computer. No cloud. No third party. Your data stays yours, and it all comes to you, not the other way around.
In May, the USPTO rejected all twenty claims.
The examiner’s argument, boiled down: other applications already sounded similar — including one from NVIDIA. Stack a couple of them together, he said, and a person skilled in the field could have arrived at my invention. Obvious. Rejected. Ouch.
Here’s the part that stung. Those applications were filed just weeks before mine — and published months after. They were sitting invisible in the USPTO’s queue when my attorney ran his prior art search. There was no possible way to find them. The patent world has a name for this: secret prior art. You can do everything right and still get blindsided by documents nobody on earth could have shown you.
If you’ve ever led an organization through change, you know this feeling. You did the diligence. You checked what you could check. And the thing that hits you comes from somewhere no amount of preparation would have reached.
What actually won the argument
At a high enough altitude, every AI system sounds the same. Mine and NVIDIA’s used a lot of the same vocabulary, and the examiner was pattern-matching on the words. The words matched.
My first instinct was to argue that the inventions had completely different purposes — and they did. But purpose doesn’t win patent arguments, and it doesn’t win business arguments either.
What won was descending from altitude. We stopped saying “our approach is different” and instead named the precise mechanism at the heart of the invention — the specific thing it does, step by step, that nothing the examiner cited actually teaches. Once the argument moved from high-altitude vocabulary to down-in-the-dirt mechanism, the pattern-match fell apart. The examiner ran a fresh search across every database he had, found nothing that did what we’d described, and allowed all twenty claims.
Why I’m telling you this
I watch companies lose this exact argument every day.
When a manufacturer tells me “we compete on quality” or “our service sets us apart,” they’re arguing at altitude — and at altitude, they sound identical to every competitor pattern-matching on the same words. The customers, like the examiner, reject it. Not because it’s false, but because it isn’t specific.
The companies that win are the ones that can descend: here is the precise mechanism by which we’re different, here is the measurable thing it produces, and here is why the alternative doesn’t do it. That level of specificity isn’t marketing polish. It’s earned. You get it by understanding your own operation deeply enough to name what’s actually happening inside it — which, not coincidentally, is what good data does for a business. It replaces “we think we’re better” with “here is exactly where, and by how much.”
Two other lessons from the process, free of charge:
Getting rejected next to NVIDIA is not the insult it appears to be. It means you’re working on a problem serious enough that the biggest players in AI are working on it, too. If your ideas never collide with anyone, worry about that instead.
The blindside doesn’t excuse you from responding well. The secret prior art wasn’t my failure, but it was still my problem. It didn’t matter whose fault it was. What mattered was whether we could articulate our position clearly enough to resolve it. Most organizational setbacks work the same way.
The good news: patent has been allowed and will issue in the coming months. The invention will keep doing what it was built to do: putting AI to work on personal data without surrendering it. And the lesson goes with me into every engagement — because the difference between sounding innovative and knowing exactly what makes you different is everything.
Erik Henry is the founder of Simplified Data Solutions. If you’d like help figuring out what precisely makes your operation different — and proving it with your own data — schedule a conversation ->