tidypress
May 22, 2026 3 min read

Agents and markdown

If the site lives in git, an agent is just another author.

The interesting shift with coding agents is not that publishing got easier. It is that authorship got cheaper. You still need a point of view, a place on the web, and a build step you trust. The agent did not remove those. It removed the friction between “I should write this down” and a file diff.

TidyPress is built for that world, but it is not built for agents. There is no chat surface, no draft host, no sync API. The site is whatever is in the tree after you merge. That is the same contract vim users had twenty years ago, and it is the right contract for agents too: they propose edits; you keep the repo.

The tree both of you edit

I like that boundary. Platforms that own your drafts eventually own your audience. A folder of markdown plus a static build/ output is boring on purpose. Boring survives acquihires, model upgrades, and the next hot editor.

  • docs/
    • tidypress.config.ts
    • tidypress-context.md
    • src/
      • content/
        • writing/
          • agents-and-markdown.mdx
          • hello-world.md
        • projects/
          • highlight.md
        • docs/
          • getting-started.mdx
    • build/
      • index.html
      • pagefind/

Presets exist so the tree already looks like a public site — writing on the feed, work on the home page — instead of a generic content/ dump. Before a big refactor, exporting a snapshot of titles (tidypress context) gives the model memory without giving it custody.

Same pipeline, different keyboard

So what changes when Claude or Cursor writes half the posts? Almost nothing operationally. A new essay is still a dated file under writing/. A project card is still frontmatter and a link. Reference pages land in docs/ when the site actually needs a manual — not because every tool secretly wants to be a wiki. You run tidypress dev, then tidypress build, then you ship build/ the way you always would. The agent does not get a special lane.

The part worth teaching models is structure, not syntax. After that, the pull request is the editorial layer. Yours.

Meme: one button labeled merge agent PR at 2am
New school: the essay arrived fully formed. Old school: you still click merge (and maybe fix one heading).

Human speed, agent speed, same history

I do not think “human writer” and “agent writer” are different products. They are different speeds on the same git history. Old school is typing the file yourself. New school is reviewing a diff that arrived fully formed at 2 a.m. Both end as HTML you can host on nginx, S3, or a cardboard box under your desk. The web does not care who held the keyboard.

It's not much but it's honest work meme
No agent API. No draft database. Just files in, `build/` out.

What I want from a tool here is narrow scope: stay out of the way once the markdown is saved. No second database. No publish button that only works inside a dashboard. If your agent can respect a folder layout, TidyPress will build it the same way it builds yours.

That is the whole bet. Files in, site out. Everything else — including intelligence — stays upstream of the commit.