JOURNAL · ESSAY
Why I'm Building CurXor — The Harness Is the Product
Thought derives intelligence. Models commoditize. The product is the harness that runs you — with a desk crew deriving work on metal you own.
What derives intelligence?
I led the last essay with a simple cut: take away the A. What is intelligence?
The next question is harder — and more useful for builders. What derives it?
Thought. Distillation through a life. Books, culture, the open web, lived experience — absorbed, transformed, kept. Through all of that, the profile spine of who you are remains.
Labs will keep arguing about thinking. Distillation will keep showing up as the word of the week. I'm not here to settle that debate. I'm asking where the intelligence an operator actually uses comes from — and whether it should belong to a universal thought-leader, or to a person.
It should be personal. That is why I am building CurXor.
Models are weather. The harness is the product.
Models get cheaper, better, and more interchangeable. That is not a crisis. That is weather. Open weights make intelligence more scalable over time. That matters. It is still weather.
The scarce layer is the harness — the applied system outside the checkpoint: memory, profile, tools, approval, jobs, scar tissue. Your rules. Your taste. Your style. The words you choose. The memory files of you. Better models make that layer more ambitious, not obsolete.
The harness is the product. The harness runs you — with a desk crew deriving work on your behalf.
Someone close to an operator put it as "AI is a better version of yourself." I would tighten that. Not a better version of you. You — with more capability.
The desk that absorbs the swap
Six weeks into shipping CurXor, we asked a simple question on the box: is our local brain stale?
The reasoning model was a generation behind. The inference runtime was a few versions behind. We updated both on the same MS-S1 MAX.
The desk adapted. Work stayed local. Still yours.
If removing one model breaks your system, you didn't own a desk. You rented a brain. You don't have a place you come back to.
We swapped. Everything else remained. Nothing skipped a beat. Thought intact. Memory intact. Models churn. Ownership is the durable layer.
Access without ownership is still rent
That is the control test I care about. Frontier checkpoints are perishable. Architecture you own survives the swap. The applied layer — memory, profile, tools, approval — does not rewrite itself every time a lab ships a new release. It grows as you do.
The defining question of the age isn't whether powerful AI will exist. It's who gets access — a few institutions, or a tool that can empower everyone. I agree with that frame. I would add the clause operators feel every day: access without ownership is still rent.
Homelab culture already proved the rent-versus-own math for people willing to duct-tape. CurXor ships the appliance desk — not a tutorial.
Why CurXor now
Because the stack is moving at speeds that punish rented-only setups. In early sessions with operators, the enthusiasm was not for a model card. It was for what this could be: what I could use it for, how I store it, being offline, keeping it on my paper.
Naval still lands: you can do almost anything — you have to know what you want to do.
Desk crew
That is what a desk crew is for — digital employees, jobs with gates. Each crewmate owns a lane. They live in Crew Cafe — a home for the work on your box, not another chat tab. They let you explore what can be done while you stay the principal. Local by default. Bring your own keys when you need more intelligence for a turn.
The offer I'm almost willing to say out loud: a one-time payment for hardware that is capable and software that delivers — about four thousand dollars — your personal algorithm on metal you own. Not another tab you rent forever.
Purpose
Purpose, for us, is simple: your personal algorithm — built with your companion AI — that you own. Not universal. Extremely personal. You with more capability. Aligned with how you think and what you care about.
You, extended
Back in 2005 I started writing a book — Over Time — about execution extended at light speed: arrive anywhere with anything. Still being written. The players changed. The need did not.
I am building this for a life that has to last — not for a demo week. In June I named the commitment in public, then stopped waiting for perfect conditions and started shipping the box. Weeks in, I am living it — not as a pitch deck. As desk time.
The slogans say: if you don't use AI, AI will use you. I don't agree with the surrender in that line. We need our own AI. That is what this box is for. You stay the principal.
We are still early. Pre-revenue. Real metal. Real operators starting to sit down. The brain can swap and your thought stays intact. That is why I am building CurXor.