So, here’s something that I didn’t expect to be saying in 2026: There seems to be a nonzero chance that Markdown might become the new RSS.

“Whoa, crazy talk! It’s not even a protocol!” I hear you saying. But the evidence has seemed to pick up of late in a couple of different directions.

The first is the budding interest in publishing on the AT Protocol, which is working to solve the network-effect challenges that have forced many of us to send newsletters rather than post blogs on RSS feeds.

That’s exciting, if incredibly niche. But simultaneously, massive developer platforms are starting to offer something called “Markdown for Agents”—something Cloudflare announced late last week, and which Laravel Cloud quickly followed up on a few days later. And Vercel jumped on it a couple of weeks ago.

(The news wasn’t all good for Markdown, but most of it was.)

Some SEO old hands, like my friend Jon Henshaw, have reacted to this news with skepticism, having had bad old memories of Google AMP and its sibling technologies Signed Exchanges and Core Web Vitals:

It’s 2026, and now I’m reading everywhere that all our pages must have Markdown versions, and it feels like AMP (and SXG and CWV) all over again. Except this time, the promise is that AI agents will better understand and interact with your site if you have them. The rationale is that HTML is too complex and consumes too many tokens to parse and analyze content. Whereas Markdown pages, with their simplicity, are ideal.

(Side note: Core Web Vitals make me want to pull my hair out.)

Jon is a smart guy and follows this stuff closer than me (Coywolf News is a great site), but I will casually defend this push towards Markdown as a lingua franca of the Web. (Not the agentic Web. Just the Web. More on that later.) I actually think it’s really a great move for publishers that comes with way fewer inherent issues than Google AMP ever did.

For one thing, this is all standards-based, not something that was just invented that you need to manage. It’s literally using existing content negotiation headers that web servers already support, not forcing folks to learn something new. Plus it’s hard to argue with a point like this from Vercel:

A typical blog post weighs 500KB with all the HTML, CSS, and JavaScript. However, the same content as Markdown is only 2KB. That’s a 99.6% reduction in payload size.

That’s good for budget-minded AI agents, but it’s also good for people who run websites.

Additionally, Markdown has been in increasingly wide use for 20 years, and it keeps growing in popularity—and unlike the weird carousels and oddly specific rules of Google AMP, lots of people know how to use it. And the use of headers to deliver Markdown pages is already baked into Web standards, just waiting for folks to use it.

OK, so how many of these servers are getting flooded with request from AI agents right now? (DepositPhotos.com)

This is really a tactic to help site owners avoid an AI-generated hug of death

Plus, there’s the rendering—Markdown is an antidote to the internet we currently run, which is highly dependent on programming languages and visual tricks that AI agents and honestly most people don’t even need. To me, when I see, “Cloudflare wants to give every webpage a Markdown version,” my thought is essentially, “Oh, they want to make AI agents stop DDoSing these poor PHP servers that still dominate the internet.”

When I see publishers talking about how their sites are getting flooded with viewers and getting slammed with unwanted hosting bills, it is clear that what we are doing is not tenable. Having Cloudflare put up a static Markdown file that takes up less space and has 0% of the JavaScript of the main page sounds like a win to me.

And if you’re building your pages semantically, as many publishers are likely already doing because they want to rank on Google, converting all that content to Markdown is going to be a cinch. Frequent Tedium skepticism target Matt Mullenweg is pushing for its addition to the WordPress.org website.

Just imagine, if you’re running an open-source project, and you didn’t have to force your users to see a loading page with anime characters just to keep the site online. Instead, you could tell Claude and Gemini and Perplexity to grab the data in a format they already use, and serve that in a static form, saving your poor forum from being drowned in dynamic requests.

There are lots of ethical qualms with AI, and you may want to just block them entirely, as is your right as a site owner. But I think diminishing a new-every-load HTML page to an unchanging Markdown file could save a lot of processing cycles for legacy server owners who have been trying to keep an extremely popular wiki online for 20 years.

I think there are websites and forums out there that have been absolutely wrecked by the rise of AI. Cloudflare, while still facing periodic reputational issues, has offered itself up as a line of defense for publishers. That’s noble—and while I get not everyone likes them, I think this particular offering is a good-for-the-internet move long-term.

And while we’re at it, let’s start printing books in Markdown. Yeah! Let’s really take this idea to an extreme! (DepositPhotos.com)

But hear me out: What if we just offered our pages in Markdown because it made the internet more accessible?

Yes, the reason for all of this is AI, because everything is about AI right now, but honestly, it would be a really awesome thing to offer for regular users, too.

Recently, I’ve been trying to take on a project with the Tedium website—it’s not quite done yet, but I’m trying to get the whole thing onto the AT Protocol, mimicking my upload of my Twitter archive to Bluesky. (I’ve gotten the upload to work, it’s just the details that need to be tweaked. Here’s a sample post that came out okay.) I’m using a tool called Sequoia, which makes it possible to plug a static site into the same protocol Bluesky uses. It parses the roughly 1,300 pages and then uploads them to a server on the ATmosphere.

Genuinely cool to see this kind of collab happening in the ATmosphere.

At the center of this is something called Standard.site, which aims to make a space for long-form content on the AT protocol. It’s not prescriptive to Markdown, though you could use it to share posts in Markdown if you wanted. It sounds promising—and like the budding efforts in the fediverse, it aims to make content easier to discover. Which is the problem RSS hoped to solve a quarter-century ago, admittedly—but this is doing it with more glue.

To me, I see a connection between the push to make Markdown an undercurrent of the agentic Web and this weird experiment on the fringes of emerging social tech. And honestly I would not be surprised if web browsers plugged into these AI-targeted Markdown feeds to give users a lightweight experience. (You know what else could use this?!? Email.)

It’s so fascinating, seeing this thing I’ve come to really appreciate as a writer turn into this ad-hoc building block of the modern internet. Even if I find it uncomfortable that AI is the vessel it rode in on.

When I found it, it was my superpower—the tool I used to plow through five articles a day at a new job. It was the cruft-buster, the starting point, the README file. And now it’s become something else entirely—something that could get us back to basics without the extra cruft of AMP or the stress of Core Web Vitals. (And even better, that didn’t come from Google.)

Honestly, I’m kind of here for it.