The blank page problem
Lee Robinson removed the CMS from cursor.com and replaced it with markdown files managed by AI agents. Chris Coyier pointed out he didn't actually remove the CMS. He moved it. They're both right. And they're both missing the more interesting question.
The web is having a real conversation about whether AI makes the CMS obsolete. It's worth having. But it's focused almost entirely on plumbing. Where the files live. How agents access them. What breaks when you have two hundred pages and five editors. I think the interesting shift is happening somewhere else.
The publishing problem is solved
The arguments are familiar by now. Robinson's team at Cursor found that the CMS abstraction made it harder for agents to work with their site. They moved everything to markdown in the repo. 344 agent requests, three days, done. For a team where every author is also a developer, it made sense.
Coyier's response was sharp: "You didn't rip out a CMS, you ripped out a cloud database." The needs didn't vanish. They moved.
Both are making reasonable cases. But notice what the entire conversation assumes: the hard part of running a website is the infrastructure. Where it lives, how it deploys, who has access.
The blank page
I've watched this scene play out with half a dozen marketing teams. A person opens the CMS. The editor is polished, the pipeline works, everything is ready. And they freeze.
Not because the tool is bad. The fields are clean, the preview is live, the publish button is right there. The problem is that the tool gives them a perfect place to write and zero help knowing what to write. So they open another tab and start googling "what should we write about."
This is the moment nobody in the AI-vs-CMS conversation is talking about. Robinson's team are all developers writing technical docs. They know what to write because the product tells them. Coyier is a solo practitioner with decades of ideas queued up.
None of them are sitting in a Monday meeting with a marketing team trying to fill a publishing calendar for next month.
The CMS industry spent twenty years solving the last mile. How to get words from an editor to a web page reliably, with version control and permissions and previews. That problem is genuinely solved.
But nobody built for the first mile. The moment before writing starts. The blank page.
What a CMS should become
We've started building this into Strife. Pull in search queries and visitor data before someone opens the editor, and the starting point changes. Not a dashboard or a report to interpret. Three suggestions grounded in what their audience is actually searching for and where the gaps are in what they've already published.
The person who used to freeze picks one and adds their own angle. Or disagrees with all three and writes something else entirely, because now there's something to push against. We use OpenPanel for the analytics that power this, and the data gets more useful the longer it runs.
The blank page isn't blank anymore. The starting point shifted from "what should I write?" to "which of these is most interesting to me right now?"
This isn't about AI writing for you. We've seen what that produces: pages that rank but say nothing, words optimized for algorithms with no opinion behind them. That approach has a ceiling.
The interesting shift is AI showing you where the gaps are between what your visitors search for and what you've published. Not generating text. Generating direction.
If AI handles the mechanical parts of publishing, the CMS doesn't become irrelevant. It becomes something it probably should have been all along. A tool that helps you figure out what's worth saying.
What's worth solving
The infrastructure question matters. Robinson and Coyier are asking real things about how it should evolve.
But the question I keep coming back to is not where words live. It's how we help people know what's worth writing in the first place. That's the blank page problem. And that's where the interesting work is.