Markdown Is the Best File Format That Nobody Respects
Everyone writes Markdown. Almost nobody stores their knowledge in it. That gap is more consequential than it looks.
In 1990, a consultant specializing in legal software estimated that 70% of American law firms ran on WordPerfect. Not a niche preference — the industry standard. WordPerfect was reliable, powerful, and purpose-built for the long, precisely formatted documents that legal work demands. The firms that chose it made a reasonable decision.
Those firms still have some of those documents. Not that it helps much. The .wpd format was never publicly specified. Other software had to reverse-engineer it over three decades. A file from 1992 uses different internal encoding than one from 2004. Open one today and you'll get garbled output, a failed conversion, or a very old copy of Word trying its best. The content exists. The context is mostly gone: the formatting, the structure, the exactness that made WordPerfect worth using in the first place.
The point isn't that WordPerfect was a bad product. It wasn't. The point is that the format was the product. When the product lost market share, the format became a liability.
The pattern
This sequence repeats often enough that it has a shape. App launches, people build habits, years of notes accumulate. Then the company pivots, gets acquired, raises prices, or just falls behind. You hit export. The export is imperfect: formatting doesn't survive the translation, images land in folders with broken references, metadata disappears. You move on, slightly less sure of what you lost.
Notion stores documents as proprietary blocks. The Markdown export drops colors, callouts, synced blocks, and embeds. Images land in folders named with 32-character UUIDs. Databases export as CSV, losing every filter, relation, and rollup you configured. The export button exists and technically works. What comes out isn't quite your notes.
Evernote's ENML format is XML wrapped around HTML. Apple Notes stores data in a SQLite database inside ~/Library that's not meant to be touched. Bear uses its own format. The specifics differ; the pattern holds. The notes are yours until you want them back, at which point you discover the format was doing more work than you realized.
Why a .md file is different
John Gruber released Markdown on December 17, 2004. The design goal was stated plainly in the first project README: a Markdown document should be readable as-is, without looking like it has been marked up with tags or instructions. A # is a heading. An asterisk means emphasis. You could send it in an email and the recipient could read it without running any software.
That's not a modest advantage. A Markdown file written in 2004 is readable today in any text editor on any operating system with no installation required. It'll still be readable in 2046. Not because Markdown was engineered for permanence specifically, but because it committed to the format that already had it: plain text. Plain text has outlasted every proprietary format it has competed against. It predates every app that currently stores your notes.
The format has no lock-in by construction. There's nothing to lock.
The strange part
Markdown won. It's the default for GitHub READMEs, API documentation, developer blogs, static site generators, technical wikis. If you build software professionally, you probably write Markdown every day without thinking about it. You use it because it works: portable, version-controllable, diffable, writable in any editor, readable without rendering.
And then you open a notes app and put your personal knowledge somewhere else entirely.
The disconnect isn't irrational. Good Markdown-native tools are fewer and more demanding than polished database-backed apps. Notion is genuinely capable. The UX gap is real. But the tradeoff being made is usually invisible. You're not just choosing an interface. You're choosing where your knowledge actually lives, in what form, and what it will take to get it out later.
Most people don't think about this until they need to. By then, the export is less useful than they expected.
What format-first actually gives you
When a tool stores files as actual .md files on disk — readable in any text editor, portable as a zip, openable in twenty years without a server request — the relationship to your knowledge changes. You can grep it. Put it in git. Inspect it directly. The app becomes a layer over data you actually own, rather than a system the data is trapped inside.
This matters more when AI is part of the workflow. An AI that can read and query your knowledge base is more useful than one that can only skim it. But for that to work, the knowledge has to be in a format the AI can reliably parse. A proprietary block format that approximates Markdown on export isn't quite the same thing. The approximation is fine until it isn't.
Structured Markdown — plain text at the surface, with typed data underneath for people, tasks, and preferences — turns out to be a reasonable answer to both problems at once. Human-readable without the app. Structured enough for AI to work with cleanly.
The small bet
The law firms from 1990 weren't naive. They chose the best tool available. What they didn't ask — what most people don't ask — is where their documents were actually stored, in what format, and what it would cost to get them out later.
Choosing Markdown as your storage format is a small, boring bet. A bet that text is text, that a file you can open in any editor is more durable than one that requires a specific server, and that your notes from today will still be fully readable in twenty years without a conversion tool.
That bet has been right for fifty years of computing. The boring choice usually is.
Asgeir Albretsen is the founder of Harbor.