@AYi_AInotes: A judgment that might get me criticized. Over the past decade, we've been spoiled by Markdown, thinking it's the ultimate form of content delivery. But last night, after browsing the open-source project html-anything — which has been online for just 7 days and already has 3.3k stars — I suddenly realized something: the true destination for output formats in the AI era is...

Here’s a take that might get me flamed.

Over the past decade, we’ve been spoiled by Markdown, thinking it’s the ultimate form of content delivery. But last night, after digging into the open-source project html-anything—which launched 7 days ago and already has 3.3k stars—I suddenly had an epiphany: in the AI era, the true endpoint for output formats should be HTML.

This project comes from nexu-io, the team behind the 40k-star Open Design.

What it does is simple: You throw any content into it (MD, CSV, JSON, notes, whatever), Hit ⌘+Enter, and a local AI agent generates production-ready HTML, then one-click post to WeChat, X, Zhihu, or download as HTML/PNG.

The whole process: zero API keys, fully local, streaming rendering, interruptible anytime.

But what really hooked me wasn’t those features—it’s the design philosophy.

There are 75 Skill templates, each hard-coded with strict constraints: 8px grid, CJK font stack, contrast ≥4.5, must use real data, no lorem ipsum allowed.

These might seem like creativity killers, but they’re actually the only way to pull AI out of the slop swamp.

Unconstrained AI just freestyles forever—flashy nonsense, broken layouts, font chaos. Constrained AI, on the other hand, cranks out stunning work, because you’ve drawn the boundaries of “design discipline” for it.

This lines up perfectly with my biggest current take on AI: what it needs most isn’t freedom—it’s a proper set of shackles.

Read on, and three counterintuitive points start crystallizing.

First, older formats are actually more advanced in the AI era.

HTML dates back to 1993, but it’s web-native, inherently responsive, supports animations and dark mode—AI’s strongest suit is cranking out HTML/CSS/JS.

Markdown, by contrast, becomes the bottleneck—you still have to typeset, screenshot, export, adapt to platforms; it’s all friction.

Second, ditching cloud APIs actually boosts productivity.

Zero API key sounds like just saving cash, but once you use it, you realize it’s pure freedom: unlimited calls, zero latency, streaming interrupt without wasting tokens. If you’ve logged into Claude or Cursor locally, it even reuses the session—marginal cost $0.

So the real pros aren’t always chasing the priciest models; they’re mastering the leanest workflows.

Third, the true moat is the lightweight nature of the delivery medium.

A few hundred KB of HTML vs. a 2MB docx—that’s a 10x difference in load speed on WeChat or mobile. Everyone’s grinding on content quality, but few realize—how heavy your delivery medium is dictates how low your open rate tanks.

Back in the day, content delivery was Markdown → typeset → screenshot → tweak format → send. Now it’s ⌘+Enter → HTML straight to clipboard → send—everything in between slashed.

That’s when it hit me: html-anything isn’t really a doc tool; it’s the content production operating system for the Agent era—it pulls the endpoint of creation from draft to finished product, nuking delivery costs to zero.

Seriously recommend this—every content creator should clone and fire it up right now, pick a Skill like deck-swiss-international or doc-kami-parchment, and you’ll instantly feel how content can be shipped like this!

GitHub link: http://github.com/nexu-io/html-anything…

Are you still using Markdown for final delivery? Or have you already switched to HTML? Lately, I’ve been feeling more and more that Markdown is just a draft format. I want to hear your real experiences, folks. Drop a comment and chat!

To run it, just three steps:
git clone → pnpm install → pnpm dev
Open localhost:3000, pick a Skill, hit ⌘+Enter. Your first impression will happen within 30 seconds.

HTML doesn’t need editing. If you’re going back and forth editing content, don’t use HTML.

Really strongly recommend this—every content creator should now clone and run it. Pick a Skill like deck-swiss-international or doc-kami-parchment.

Does WeChat not accept it? Does it block it or something?

Previously I did it manually, but it was exhausting. My current workflow is: AI helps me search and organize material, then I dictate using voice input.

That’s a good method. For complex stuff, I suggest using Cloud Opus 4.7 for a one-shot output.

Exactly.

Haha, class monitor, this sentence is a must-study.

You decide! Haha, brother agrees?

Why are you so sure no one has ever said that before? Haha.

Yeah, Markdown is just a draft format.

Haha, thirty years east, thirty years west—times change.