The funeral was held on July 1, 2013. Google Reader had been announced as scheduled for shutdown on March 13, in a Google blog post titled "A second spring of cleaning," for the stated reason that "while the product has a loyal following, over the years usage has declined." Petitions were filed. Articles were written. The general consensus, in the spring and summer of 2013, was that this was the end of an era — that the open syndication web had been on life support for years and Google had finally pulled the plug.
The format that the funeral was for is now, twelve and a half years later, the substrate that every podcast, every newsletter platform, every static-site generator, every Mastodon and Bluesky bridge tool, and an increasing fraction of AI-agent infrastructure runs on. RSS in 2026 is, by every signal worth looking at — podcast feeds, WordPress installs, newsletter platforms, federated-social-web bridges — more deployed than it was the day Google Reader shut down. Twelve and a half years of every successive content platform announcing itself as the RSS replacement have produced an ecosystem in which RSS is the boring substrate underneath the platforms.
I want to walk through how that happened, because the shape of the answer turns out to be the shape of the answer for protocols-without-owners in general, and the answer has implications well past whatever happens to Twitter or Substack or AI agents next.
The format that nobody really invented
The ancestry is unromantic. The first RSS-shaped thing was released by Netscape on March 15, 1999 under the name RDF Site Summary. Its lead author was Ramanathan Guha, and the format was meant to feed Netscape's My Netscape portal — the kind of personalised home page every web company in 1999 thought was going to be the future. Within four months, Netscape released a simplified RSS 0.91 with the RDF parts removed and the file-format philosophy more pragmatic. Dan Libby authored that one. By June 2000, Netscape had lost interest entirely; UserLand, Dave Winer's company, picked up the spec, kept the same version number, and quietly continued maintaining it.
The format then forked. In December 2000, an RSS-DEV Working Group — including a fourteen-year-old Aaron Swartz alongside Guha and several others — released RDF Site Summary 1.0, which restored the RDF parts on the grounds that they were architecturally important. UserLand released RSS 2.0 in August 2002 with the RDF parts removed again, on the grounds that they were architecturally inessential. The two camps disliked each other intensely. The IETF eventually got involved and, in December 2005, published RFC 4287, defining the Atom Syndication Format as a third option meant to settle the disputes through a formal standards process. Atom is technically cleaner than either RSS variant; almost nobody outside the standards-process people actually uses it in preference to RSS 2.0.
Reading this history with later knowledge is uncomfortable. Aaron Swartz at fourteen, co-authoring a specification that would still be in heavy use a quarter-century later, before either his brilliance or the ending of his life were apparent. Dave Winer's UserLand, a small company that quietly carried the format from Netscape through the dot-com bust into the era of personal blogging. The standards process resolving nothing because by the time the formal RFC was published the format had already won by being already deployed. The whole thing is the kind of fundamentally-uncoordinated process that the open-source world produces by default and that most contemporary product organisations would be incapable of replicating if they tried.
The format itself is unceremonious. It is XML, with a tag for the channel and tags for items, and the items have titles, links, descriptions, and dates. There is no specification of who must offer it, who may consume it, what fields must be present, what user-agent must request it, or how a reader application must render it. The format is a contract about what an XML file looks like, and that is the entirety of the contract. There is no committee, no roadmap, no quarterly release schedule, no pricing tier, no auth flow, and no roadmap for deprecation. It is, by infrastructure standards, almost invisible.
The killers, in order
Every five years or so, a new content platform emerges and is described, by its founders or its admirers, as the thing that finally replaces RSS. The list of these is long enough now to be its own historical artefact.
Twitter was the most famous. Around 2008–2010, the consensus framing in tech press was that Twitter — real-time, social, mobile-first, algorithmic — was the format-shaped successor to feed readers. Twitter's API supported RSS and Atom output for roughly seven years, from the early-2006 platform through to the API 1.1 cutover in March 2013, when Twitter dropped RSS/Atom in favour of JSON-only and cited low usage. The replacement narrative was internally consistent up until the moment Twitter, having spent the next decade making itself maximally hostile to anyone wanting machine-readable timeline access, became X, restricted its API to corporate licensees, and the people who had been using Twitter as their feed reader had to find something else.
Facebook News Feed was the second. The argument here was algorithmic-personalisation: News Feed would learn what you wanted to see, and a chronological-feed format would feel quaint by comparison. The argument was made through 2010–2015 and became less convincing every year as Facebook's algorithmic curation revealed itself to be optimising for engagement rather than for what readers actually wanted to read.
Apple News came next. The argument was premium curated content with native iOS integration. Apple News shipped with RSS import on day one, in September 2015, and then removed RSS support roughly six months later when Apple News Format went live in March 2016. The launch-day decision is one of the more telling artefacts of the era — the team building the RSS replacement included an RSS reader as a core feature on day one. The removal told its own story: Apple News could only get to its preferred premium curated model by closing the open-web entry point that had brought it the initial publisher base.
Newsletter platforms (Substack from 2017, Beehiiv, Ghost) were the most recent killer. The premise was that email is the actual reading interface and the open-web syndication problem was solved by letting writers email readers directly. This is an argument that survives partial scrutiny — newsletters do work — but every newsletter platform in 2026 ships an RSS feed by default, because writers expect to be syndicatable, and "an RSS feed of every post" is the cheapest way to be syndicatable.
Algorithmic personalisation feeds — TikTok, Instagram Reels, the YouTube Home tab, every recommendation surface — are the contemporary version of the Facebook News Feed argument with a stronger machine-learning substrate. The argument is the same: chronological feeds are inferior to algorithmic curation. The track record of the argument, fifteen years in, is mixed at best, and the population of people who experimentally turn off algorithmic feeds and switch to RSS has been growing rather than shrinking through 2024 and 2025.
What's striking about reading the list in chronological order is how each killer was perfectly internally consistent at the moment of its launch. The killers weren't wrong about their own merits. They were wrong about what RSS was for.
What RSS is for
RSS is a contract that says I will make the things I publish available, in a structured form, to anyone who wants to fetch them, on a URL that doesn't change and an XML schema that doesn't surprise you, without you needing my permission or my account. That contract is a small thing in any individual deployment — a few hundred bytes of XML on a server somewhere — and a load-bearing thing across the whole web.
The killers I listed all proposed to replace the contract with a richer one. Twitter offered real-time. Facebook offered personalisation. Apple News offered curation. Substack offered direct-to-inbox delivery. Each replacement was, on balance, a richer contract — and each came with a corollary that the contract was now between the publisher, the platform, and the reader, with the platform owning the relationship. That ownership is what RSS doesn't have, and the absence is what makes the format survive.
Three populations of users have, over the last twelve years, been quietly demonstrating this in different ways:
Podcasts. Apple's Podcast Directory, from iTunes 4.9 in June 2005 onwards, accepts only one input format: an RSS feed. Spotify, when it entered the podcast market, built ingestion against RSS. Every podcast hosting platform, every podcaster, every aggregator, every transcript service in 2026 reads RSS, because Apple's directory ingests RSS and Apple's directory is where listeners find shows. The format that was supposed to be replaced is the format the entire global podcast industry runs on, twenty-one years after Apple shipped it.
Static-site generators. Hugo, Jekyll, Eleventy, Astro, Zola, Next.js — every modern static-site generator and most modern blog frameworks generate RSS by default, because the developers building the tools assume their users want their sites to be syndicatable, and RSS is the cheapest way to provide that. The default in this ecosystem is a feed exists, and most users never disable it.
Open-web publishing platforms. WordPress, still running on roughly 42% of the web by W3Techs's count, has shipped RSS by default since the original 2003 release. Substack ships RSS. Ghost ships RSS. Mastodon, which uses ActivityPub for federation, also publishes RSS feeds for any account because content portability is part of the platform's premise. The newsletter platforms ship RSS, the federated platforms ship RSS, the legacy CMSes ship RSS. Anywhere that someone publishes things and other people read them, RSS gets shipped, because not shipping it would require explicit work and the publishers don't want to do that work.
The killers replaced the relationship around content. They didn't replace the contract that says the content is fetchable. The contract is what kept being shipped, on the same XML format, with the same schema, on the same URL conventions, year after year, mostly because it's free to ship and there was no reason to stop.
What the boring protocol affords
There is a 2026 chapter to this story that's worth telling honestly, because it's both real and easy to overstate.
Over the last eighteen months, the AI-agent ecosystem has been rediscovering RSS, in the same way every previous generation of new tooling has rediscovered it. The Model Context Protocol now has a community-maintained feed-mcp server that exposes RSS, Atom, and JSON Feed to LLM-based agents. RSS3, a Web3-flavoured project, launched AgentData on December 4, 2025 — explicitly framed as "real-time internet feeds built for AI agents." Every workflow-automation platform — n8n, Zapier, Make — now has documented templates for routing RSS items into Claude or GPT or Perplexity prompts. Specialist content-monitoring agents pull RSS where it exists, scrape HTML where it doesn't, and treat RSS as the higher-quality source.
What the agent ecosystem has not done is save RSS. RSS didn't need saving. The agent ecosystem walked into a room where the format was already on the table and noticed that it was easier to read than the HTML around it. The framing in some of the recent coverage — AI agents have given RSS a second life — gets the causality backwards. RSS gave the agents a second hand-hold on the open web. Without it, every agent's content-ingestion pipeline would have had to do its own HTML scraping, which is approximately what they do anyway when a site doesn't expose a feed, with the predictable result that the unfeeded sites are harder to ingest cleanly than the feeded ones.
The interesting thing about that asymmetry is that it's an end-state, not a transition. The HTML the agents scrape is going to keep changing as web frameworks change. The RSS feeds the agents read are going to look almost identical to what they looked like in 2005, because the format has had no incentive to change and no owner empowered to change it. The agent ecosystem, which moves at terrifying speed in every other axis, is going to find that the structured-content layer it's reading from is the most stable thing in its stack.
The list of RSS killers is itself worth a quick walk. Twitter, from roughly 2008 to 2013, was supposed to be the real-time social timeline that made polled feeds obsolete; it rate-limited and then removed RSS export, made itself hostile to feed-style usage, and RSS survived. Facebook's News Feed, between 2010 and 2015, was the algorithmic-curation version of the same argument; it optimised for engagement rather than for what readers wanted to read, readers noticed, and RSS survived. Google Reader's 2013 shutdown — the canonical RSS funeral — was supposed to be the end; hosted readers like Feedly, Inoreader, and Newsblur proliferated in the immediate aftermath, and what people had really depended on was a reader, not the format. Apple News, from 2015 through 2020, was the premium-curated answer; the product shipped with RSS import on day one in 2015, Apple News Format went live in March 2016, RSS support was then removed inside the product, and RSS continued unaffected in every other publishing tool publishers were already using.
The pattern continued. Newsletter platforms, from 2017 forward, were email as the reading interface; every newsletter platform shipped RSS by default because writers expected to be syndicatable. Algorithmic personalisation feeds, from 2018 forward, were the chronological is dead argument; algorithmic feeds do dominate consumption time today, but chronological RSS persists for the population that actively prefers it, which is growing. And the AI-agent ecosystems, from 2024 forward, are the latest agents replace browsing framing; agents read RSS where it exists, which means the killer is the killer that depends on the ostensibly-killed format.
What the boring protocol's survival actually shows
The temptation, reading this list, is to draw a moral lesson. Centralised platforms always collapse; decentralised protocols always survive. That isn't true. Most decentralised protocols of the 2000s did die. Usenet died. Jabber/XMPP died as a public-facing thing despite being technically cleaner than what replaced it. The IRC fediverse limped along until Slack and Discord finished it off. RSS isn't the rule; it's the exception, and the exception is worth understanding because the conditions that produced it are not magic.
What kept RSS alive, on closer inspection, is a combination of properties. The format is small enough that a publisher's RSS feed is almost free to ship, which means there's no business case for stopping. The format has no owner who could announce a sunset. The format is consumed by enough downstream tools — podcast directories, static-site generators, feed readers, email-newsletter ingesters, now AI agents — that any individual publisher would lose machine-readability across many ecosystems by removing it. The format is so simple that user-built clients keep being viable; there is no API key, no rate limit, no commercial relationship to revoke. Each property is small. The combination is the thing that's hard to kill.
Compare to the protocols that did die. Usenet had owners (the Big Eight cabal); the format was complex enough that maintaining a server was real work; the spam-defence economics inverted in the late 1990s. XMPP had a working IETF spec but no consumption-side incentive — every chat product wanted to vendor-lock its users. IRC had no spec the major networks actually adhered to, and no plausible commercial relationship for anyone to support it through.
RSS, by contrast, has an infrastructure-grade set of survival properties: simple, ownerless, cheaply shipped, broadly consumed, free of commercial relationship. None of these are the kind of property a venture-backed product organisation can replicate. They are the properties of something that emerged in 1999 from a portal-page feed and got slowly turned into a substrate by everyone who wanted a substrate and didn't care to build their own.
That's the part that matters past RSS specifically. Every time a contemporary platform announces a new open standard — for activity, for content portability, for inter-app messaging, for AI-agent context — it's worth checking whether the proposed standard has the survival properties or only the marketing of survival. Activity, owners, complexity, the absence of commercial relationships, the size of the consuming-ecosystem: these are the diagnostic questions. RSS is the answer in the affirmative on all of them. Most "new open standards" announced in 2026 will fail on at least three. The ones that succeed will look like the boring 1999 thing more than they look like the press release announcing them.
Coda
A small, badly-coordinated, format-fork-prone, owner-free XML schema, defined in 1999 and never seriously revised, is the substrate that the podcast industry, the newsletter industry, the static-site-generator ecosystem, the federated-social-web ecosystem, and now the AI-agent ecosystem all stand on. None of those industries set out to use RSS. Each of them found it already there, working, with no maintainer to negotiate with and no migration cost. That is the property worth carrying forward: the protocols that survive are the ones nobody had to keep alive.
The funeral on July 1, 2013 was for one product, owned by one company, that read one format. The format kept going because it was never owned by the company in the first place. That is the whole story, and it is also the only story that matters about the open web.
Top comments (0)