A computer with no operating system
No OS to boot, no applications to open, no file system to navigate. You boot into the model, and what you keep is versioned artifacts. I filed it provisionally, then built enough of it on an old Dell to know the idea holds.
The personal computer’s basic shape hasn’t changed in forty years. An operating system loads, you pick an application, the application opens a file, and you do the work inside that application’s idea of what work is. Every one of those layers was a reasonable answer to a hardware constraint that no longer applies.
The premise of the filing is that you can remove all three. Boot straight into the model. There is no application to choose, because the thing you’re talking to can do the work directly. There is no file system to organize, because what you keep isn’t a file - it’s an artifact, versioned, with the reasoning that produced it still attached.
That last part is the piece I care about most, and it’s the piece that comes from the day job rather than from theory. Once you put autonomous agents anywhere near enterprise data, the problem stops being capability and becomes provenance: which version is current, what changed, who or what changed it, and can you reconstruct the state a decision was made in. Ordinary file systems answer none of those questions. They were never asked to.
The filing
“Computer System with a Versioned, Semantically-Indexed Artifact Store as the Native User-Facing Storage Layer for an Artificial-Intelligence Operating Interface.” USPTO provisional application no. 64/103,768, filed July 2026, developed independently.
I’m keeping the mechanism off this page while the provisional is live. What I’ll say about the build: the hardware was a Dell that had been sitting unused, the point was to find where the concept breaks rather than to demo something polished, and it broke in more interesting places than I expected - mostly around what a user does when there’s no folder to put something in.
If you work on storage architecture, agent infrastructure, or anything adjacent, I’d rather talk than publish. That’s not coyness about the IP. It’s that the useful conversations so far have all been with people who’ve hit the same provenance wall from a different direction.
Read next
Something I’m working on
What does AI actually know about a business, and does it get it right? That question turned into a GEO Review, a practical look at how a business shows up in AI search.