I tried to automate my blog last month. The writing part was easy — Claude handled that in seconds. Then came the hard part: copy the markdown, open the CMS, paste it in, fix the formatting that broke during paste, manually set the meta description, upload an OG image, click publish. Repeat for the next post. And the next one.
The AI wrote the content in ten seconds. Publishing it took ten minutes. Every single time.
That's when it clicked. We don't have a publishing problem for humans. WordPress, Ghost, Medium — they're all fine if you're a person sitting at a keyboard. But the moment an AI agent enters the picture, every one of these platforms falls apart.
The gap nobody filled
Go try to publish a blog post programmatically right now. Your options are grim.
Medium killed new API key issuance years ago. Naver Blog and Tistory shut their APIs entirely. WordPress has a REST API, but you're signing up for self-hosting, plugin hell, and PHP maintenance. Ghost gives you an API too — behind a $15/month paywall and your own server. Dev.to has an API but no custom domains. Hashnode has GraphQL but zero AI automation features.
None of them speak MCP. None of them have heard of llms.txt. None of them were designed for a world where your content pipeline starts with an AI agent, not a person opening a browser tab.
There's a whole cottage industry of "AI blogging tools" — Autoblogging.ai, WPAutoBlog, Emplibot. They're all wrappers around WordPress. If WordPress changes an API endpoint tomorrow, they break. They don't solve the problem. They duct-tape over it.
What "AI-native" actually means
I want to be precise about this because the term gets abused.
Postlark is not an AI content generator. We don't write your posts. We don't call OpenAI or Anthropic on your behalf. If you want to think of an analogy: YouTube doesn't shoot your videos. Postlark doesn't write your words. You bring the content — whether you wrote it yourself, had Claude write it, or used some Rube Goldberg n8n pipeline. We don't care. That's your business.
What we do care about is making the publishing infrastructure native to AI agents. That means three things.
First, an MCP server that ships as an npm package. If you're running Claude Code or any MCP-compatible agent, you add Postlark as a tool and your agent can create, update, schedule, and manage posts directly. No browser. No copy-paste. No human in the loop unless you want one.
Second, llms.txt. This is a file that tells AI agents how to use your blog platform — what endpoints exist, what they do, how to authenticate. A non-developer can literally say "set up a blog on Postlark and start posting" to Claude, and it figures out the rest by reading llms.txt. No documentation hunting. No Stack Overflow.
Third, a CLI. Because sometimes you're already in the terminal and you just want to run a command. postlark publish and you're done.
The REST API sits underneath all of this, obviously. But the point is that the API isn't an afterthought bolted onto a web dashboard. The API is the product. The dashboard is the afterthought.
The workflow we wanted
Here's what publishing looks like before Postlark: you generate content with AI, copy it, open your CMS in a browser, paste it, manually configure SEO metadata, manually set an OG image, click publish. For one post, it's annoying. For ten posts across three blogs, it's your whole morning.
Here's what it looks like with Postlark: you tell your AI agent to publish. It does. SEO metadata, OG image, JSON-LD, RSS feed, sitemap — all handled automatically. You move on with your day.
That's not a marginal improvement. That's a category change. The human goes from being a manual operator to being an editor who reviews and approves. Or, if they trust their pipeline, they go do something else entirely.
Who this is for
Developers running automated content pipelines. Indie hackers who blog about their builds. DevRel teams pushing out technical content. Anyone who's looked at their publishing workflow and thought "why am I the bottleneck here?"
If you write one personal essay a week in your browser, WordPress is great. Stick with it. Postlark isn't trying to replace that.
But if you've got an AI agent generating content and you're still manually ferrying it into a CMS — that's the gap. That's the thing that shouldn't exist in 2026.
What comes next
This is the first post in a series called Building Postlark. I'm going to walk through the decisions we made, the trade-offs we hit, and the things that surprised us along the way. Not a changelog. Not a feature announcement feed. The actual messy story of building a platform from scratch.
Next up: why we threw out the rich text editor and went all-in on Markdown. It wasn't obvious. It almost didn't happen. But once we understood what AI agents actually need from an input format, WYSIWYG stopped making sense entirely.