Arbiris — Deterministic Agentic Intent Infrastructure
Explaining cryptographic proof to people who will check
Arbiris is infrastructure for deterministic agentic intent — autonomous workflows whose actions can be proven, rooted in hardware. The build problem was communication before it was engineering: the audience is technical, sceptical by profession, and will read the claims closely.
- Type
- Technical platform & content site
- Role
- Front-end architecture & content layer
- Core stack
- Next.js · React 19 · TypeScript · Sanity CMS
- Focus
- Technical credibility & content velocity
The problem
Deep-technical products are usually sold to one of two audiences with one piece of writing, and it fails both. Simplify the cryptography and the engineers evaluating you conclude there is nothing underneath. Lead with the primitives and everyone else leaves before reaching the reason any of it matters.
This category makes that worse. "Deterministic agentic intent" and "hardware-rooted cryptographic proof" are precise terms carrying real claims, and the audience is exactly the group that will test whether the claims survive contact with detail. Marketing language that overreaches here does more damage than saying less.
The second problem was velocity. Credibility in infrastructure is accumulated through published technical writing, not asserted on a homepage — which means the blog is not a section of the site, it is the part that has to keep moving after launch without an engineer in the loop.
What I built
A Next.js and React 19 platform with a Sanity-backed content layer, built so the marketing surface and the technical writing reinforce each other.
A layered landing experience — the argument stated plainly up top, with the mechanism available immediately below for readers who want it.
A searchable technical blog — the content layer where credibility is actually built, searchable because reference material that cannot be searched does not get used.
Proof storytelling — the hardware-rooted proof model explained as a mechanism a reader can follow, rather than asserted as a feature.
A typed content model — the structure of a post is defined once and enforced, so publishing cannot silently break a page.
The technical decisions, and why
Progressive disclosure instead of picking one audience
Each section states its claim in plain language and then, in the same place, offers the mechanism behind it. A reader deciding whether this is relevant gets that in a sentence; a reader deciding whether it is real gets the detail without navigating away. This is what lets one page serve both audiences honestly rather than splitting into a marketing site and a docs site that contradict each other.
Sanity as a structured content layer, not a page builder
The content model defines what a technical post is — its sections, its code blocks, its references — rather than storing a blob of rich text. Structure is what makes the archive queryable later: search, related posts and topic pages all become straightforward, and none of them are possible once the content is a wall of HTML. It also means writing does not require a deploy, which is the actual constraint on publishing cadence.
Claims written to survive scrutiny
For this audience, precision converts better than enthusiasm. Copy is scoped to what the system does — the words "deterministic" and "hardware-rooted" are load-bearing and used only where they hold. Overstating to a group that will check is the fastest way to lose them, and there is no recovering the reader who caught you.
Motion kept subordinate to the reading
Long technical pages punish decoration: animation that interrupts a paragraph costs comprehension. Framer Motion is used for orientation — signalling structure and transitions between sections — and stays out of the way everywhere the reader is actually reading.
Where it landed
The site holds a genuinely difficult line: legible to a first-time reader without thinning the claims for the engineers evaluating them.
The structured content layer is the part that keeps paying. Because posts are modelled rather than pasted, the archive stays searchable and extensible as it grows — and publishing stays a writing task rather than an engineering one, which is the only way technical authority actually accumulates.