# Hello, world *Hello humans, hello agents.* Published 2026-04-28 · https://dom.vin/2026/hello-world ## Summary This is the introduction essay for the rebuilt dom.vin, by Dom Vinyard, who works on agent intelligence at sauna.ai. The site will host a small number of interactive, illustrated essays on AI, written deliberately for two readers in parallel: a human who scrolls a rendered page with figures, and an agent that fetches a markdown mirror at //.md and reads prose plus a structured paragraph wherever a figure would have been. Every figure on the site carries a required prose description, treated as part of the essay rather than accessibility decoration; if the description is missing, the build fails. A site index for agents lives at /llms.txt. The site does not chase frequency or engagement. ## Claims - dom.vin is being rebuilt as a small home for slow, illustrated essays on AI. - The site is designed for two parallel readers: a human reading a rendered page, and an agent reading a markdown mirror. - Every figure must carry a required prose description; the description is part of the essay, not accessibility decoration. - An agent index at /llms.txt lists every essay with summary and claims. --- This is dom.vin, rebuilt. A small home for essays on AI — interactive, illustrated, slow to write. I'm Dom. I work on agent intelligence at [sauna.ai](https://sauna.ai). I think most clearly when I write things down, so I'm doing more of it here. What's different: each essay is built for two readers. One of them is you — scrolling, watching figures unfold, reading on a phone in the morning queue. The other never loads the page. It fetches a markdown file and reads prose. Where a figure would have been, the markdown reader sees a paragraph that describes it. > **Figure 1 — reader-split.** At the top of the frame, a code-editor window depicts the essay's MDX source, complete with traffic-light dots, a 'essay.mdx' filename, line numbers, frontmatter, prose, and a highlighted JSX tag — the very tag that summons this illustration. Below the source, a 'renders to' divider line splits into two dashed paths fanning down to two outputs, side by side. The left output is a browser window labelled 'rendered': it shows the actual essay title in serif, real prose lines, and a sage-bordered figure box containing a miniature illustration. The right output is a file-style window labelled 'markdown mirror': it shows the same essay's frontmatter and prose in monospace, interrupted by a rose-ruled quote block holding the actual figure description. A numeric marker [1] ties the JSX tag in the source to the figure box on the rendered side and the quote block on the mirror side. The split is in presentation, not in content. Both readers get the same essay. What changes is how each figure arrives — once rendered, once written down. The written one isn't optional. A figure I can't describe in prose is one I haven't finished thinking about. > **Figure 2 — equivalent-representations.** Two windowed columns flank a central operator. The left column, headed 'figure', contains an annotated diagram: at the top, two small input boxes labelled [1] tokens (a pastel-stippled rectangle) and [2] tools (a rose-bordered rectangle) feed via arrows into [3] the agent — a halftone-tinted central rectangle that composes them — which in turn feeds via an arrow labelled 'resolve' into [4] a sage-circled endpoint marked 'output'. Tick marks line the frame edges. The right column, headed 'description', writes itself in mono prose line by line, and each numeric marker [1]…[4] maps directly to a tagged paragraph that describes the corresponding part of the figure. Between the columns sits a triple bar (≡) in sage, labelled 'equivalent representation', with thin dashed ties extending up and down from it. A 'step N/4' counter at the top of the figure column ticks 0 → 1 → 2 → 3 → 4 as the parts resolve. The figure and the description carry the same information; an agent reading only the prose arrives at the same structure as a human reading only the diagram. The whole site is shaped this way. There's a [/llms.txt](/llms.txt) at the root for agents looking for an entry point. (They're first-class readers now. On this site, at least.) What I won't do here: post often, optimise for engagement, write about anything that doesn't actually pull at me. The pieces are the questions that won't leave me alone — the ones I'd be writing whether anyone read them or not. More soon.