HTML as Native Output
The output format a model is fluent in is the one
its training data is saturated with — and for laying out text and images
beautifully, that is HTML. The framing: if Markdown is the
model's native language on the input side, HTML might be its
native language on the output side. The stated mechanism is
training-data composition — the web is full of HTML pages, so the model
is excellent at the question "how do I take a pile of text and images
and lay it out into something clear and beautiful," and emitting HTML
lets it use its full front-end design muscle rather than being boxed
into a format (like a .pptx object model) it has seen far
less of.
The payoff has a second half that makes HTML special among native formats: it is friendly to both the model and the human. Markdown is friendly to the model but not to us (nobody enjoys reading a wall of text); a rendered HTML page is pleasant for both. So choosing HTML as the output isn't just "pick what the model writes well" — it is the rare format where the model's fluency and the reader's experience point the same way.
Claims
- A model is most capable in the output format its training data is saturated with; the web makes HTML that format for laid-out text and images. principle — durable as long as models are trained on web-scale corpora: fluency tracks exposure, and choosing an output representation the model rarely saw taxes it. This is Agent-Computer Interface (ACI)'s "keep formats close to what the model has seen naturally," applied to the output channel rather than tool I/O.
- If Markdown is the model's native input language, HTML is its native output language. observation — the video's framing, offered as a guess about why the model designs so well in HTML; a plausible model-affordance heuristic, not a proven claim about the training distribution.
- HTML is friendly to both the model and the human, where Markdown is friendly only to the model — which is why HTML uniquely serves generation and consumption at once. principle — durable: it names the property that makes HTML preferable to other model-fluent formats for human-facing output; the format serves both ends of the pipe.
- Emitting HTML lets the model use its full front-end design muscle, producing output "flat-out impossible in PowerPoint" (interactivity, animation, embedded live media). observation — the capability claim behind the pitch; the ceiling is the open web's, not a slide tool's.
Related
- Agent-Computer Interface (ACI) — the input/tool-side sibling: "choose formats cheap for the model to produce, close to what it saw naturally." HTML-as-native-output is the same heuristic pointed at what the agent emits for humans.
- Golden Templates — how you steer the model's native HTML fluency toward your taste rather than generic output: curated, opinionated exemplars it adapts.
- Agentic UI Testing — because the output is a rendered web page, the agent can screenshot and self-correct it; native output and visual self-validation compound.
- Text-Space Optimization — the adjacent idea that the representation you hand the model changes what it can do; here the representation is on the output side.
- Knowledge Work as Code — HTML being the model's native output is one reason "everything is code" pays off: the code the model most wants to write is also human-friendly.
- Distillate: Stop Making PowerPoints: Vibe-Coding HTML Slides as a Skill