Netscape, Dave Winer, and the Format That Arrived Before Anyone Needed It

A standard built in a quiet room, adopted in a noisy one

In the spring of 1999, Netscape was losing a browser war it had already effectively lost. The company was being absorbed by AOL, whose acquisition closed in March 1999, its engineering staff was scattered across projects that would never ship, and its portal ambitions — My Netscape Network, a personalised homepage — needed a way to pull fresh content from partner sites automatically. The solution a small team reached for was RDF Site Summary, a format that combined the W3C's Resource Description Framework with a simple channel-and-item structure for syndicating headlines. Version 0.9 appeared in March 1999. It was built partly on ideas from an earlier Netscape format called MCF and drew on work that Tim Berners-Lee's RDF working group had been developing at the W3C. It was also, for most practical purposes, unworkable.

RDF's strength is precision: every statement about a resource can be formally related to every other statement in a machine-readable graph. Its weakness, in 1999, was that almost no one outside academic metadata circles could parse it correctly, and the tooling to do so barely existed. RDF Site Summary 0.9 sat at the intersection of an ambitious standard and an absent ecosystem. Netscape's portal team lost interest almost immediately, and the format's early momentum stalled before it had a meaningful install base.

Dave Winer's fork and the long argument that followed

Dave Winer was already well placed to notice the opening. His company UserLand Software had been building tools for web publishing since the early 1990s, and Winer had a persistent, well-documented interest in giving personal websites the same syndication capability that wire services and news portals enjoyed. When Netscape published its 0.9 draft, Winer worked with the Netscape team to produce RSS 0.91 — a version that stripped the RDF layer almost entirely and replaced the formal ontology with a much flatter XML structure. The channel had a title, a link, a description. Items had the same three fields. A developer could implement a parser in an afternoon.

RSS 0.91 was the format that actually spread, first through UserLand's Radio and Manila publishing tools and then, as the blogging wave gathered in 2001 and 2002, through Blogger, Movable Type, and eventually WordPress. Publishers did not generally ask for it; their software emitted it, and over time feed readers appeared to consume it. The infrastructure preceded the demand by just enough time to be ready when the demand arrived.

The argument about what RSS actually was — whose format, whose process, whose namespace — began almost as soon as the 0.91 draft settled. A 0.92 and a 0.94 followed, adding the enclosure tag in 2001 and a cluster of optional elements. By 2002 Winer had published RSS 2.0 under a Creative Commons licence and transferred stewardship to Harvard's Berkman Center, an attempt to put the specification beyond the reach of future corporate capture. At roughly the same moment, a separate group of developers — including Sam Ruby, Mark Nottingham, and Aaron Swartz — concluded that the RSS process was too chaotic and too personality-dependent to serve as infrastructure for the open web, and began work on what became the Atom specification, eventually published as RFC 4287 in 2005. The technical differences between RSS 2.0 and Atom are real but modest; the governance differences were the actual point.

The blogosphere that needed it

What neither the Netscape engineers of 1999 nor Winer in 2001 could have fully anticipated was the scale of the environment that would depend on their work. Technorati, founded in San Francisco in 2002, tracked links between blogs and at its peak indexed tens of millions of sites — feeds were the mechanism by which it received notification of new posts. Bloglines, launched in 2003, was among the first web-based feed readers to approach mass use, demonstrating that the aggregation side of syndication had commercial potential. Nick Bradbury, working from Nashville, built FeedDemon, the dominant Windows desktop reader of the era. The format Winer and the Netscape team had assembled between 1999 and 2002 was suddenly running an appreciable fraction of the web's information flow.

A feed is a text file sitting on a server. Between the publisher and the reader there is a fetch, and nothing else.Photo: Tima Miroshnichenko / Pexels

Google Reader, which launched in Mountain View in October 2005, was the moment the infrastructure became invisible in the way that all successful infrastructure eventually does. Google Reader's shutdown in July 2013 did not kill RSS, but it did collapse the casual-user layer of the ecosystem, leaving the format in the hands of people who specifically wanted it — developers, journalists, and readers who understood what polling an endpoint gave them that an algorithmic timeline could not.

The blogosphere's peak, roughly 2004 to 2008, is the period in which RSS's original design decisions proved themselves. The conditional GET mechanism — feed readers using HTTP ETag and Last-Modified headers to check for updates without downloading an unchanged feed — kept bandwidth costs manageable at scale. The format's deliberate simplicity meant that any CMS, however small, could emit a valid feed without a dedicated engineering team. And the fact that Winer had placed RSS 2.0 under an open licence meant that no company could charge for the right to publish or read in it.

Google Reader, which launched in Mountain View in October 2005, was the moment the infrastructure became invisible in the way that all successful infrastructure eventually does.

What the format did not include was equally revealing. There was no authentication model. There was no mechanism for a publisher to know who was subscribing. There was no click-tracking, no pixel, no open rate. A reader who subscribed to an RSS feed disappeared into their client and left no trace, which is precisely why the format found its second life — after the newsletter pivot of the 2010s made email's surveillance infrastructure visible — among readers who preferred their reading behaviour to remain private.

The accident of timing

The title "RDF Site Summary" describes a format designed for headlines on portal pages. What RSS became — the syndication layer for an entire era of decentralised publishing — was not designed. It accumulated through a sequence of practical decisions, personality-driven forks, and the accident of arriving just before the blogging wave needed it.

The people involved were not building for posterity. Netscape wanted to populate a homepage widget. Winer wanted his publishing tools to interoperate with other publishers' tools. The developers who wrote Atom wanted a process they could trust. None of them set out to lay infrastructure that would still be emitting feeds from static site generators a quarter-century later, or that would become the technical basis of a podcasting industry that Spotify and Apple would eventually fight to control. They solved the problems in front of them, and the solutions outlasted the problems.