A Consortium Is Born
The second half of the second week of December 1994 was a pivotal stretch for the web. On Wednesday the 14th, about two dozen representatives from major tech companies, browser makers, and web-based startups gathered around a table at MIT for the first meeting of the World Wide Web Consortium. The agenda was mostly administrative: the role of the consortium, how it would conduct itself, and its responsibilities to the wider web. By the end of the meeting, the group felt confident about the future of web standards.
The consortium had been on Tim Berners-Lee's mind since 1992. After years of conversations with experts and advisors, he left CERN in October 1994 to run the organization at MIT's Laboratory for Computer Science. He had no intention of playing dictator, and he preferred to listen. The model he envisioned was one of compromise: shared technical documentation that competing companies could orient themselves around, creating interoperability through consensus rather than through market dominance.
The next day brought a milestone of a different sort. Jim Clark and Marc Andreessen announced the wider release of Netscape Navigator 1.0, which had been in beta for months. Initially given away for free in a bid for market share, the browser was a commercial and technical success, improving on the speed, usability, and features of its predecessors.
On Friday the 16th, the W3C hit its first setback. Berners-Lee had planned for CERN to serve as the European host of the consortium, but that day CERN approved a massive budget for the Large Hadron Collider, leaving little room for hypertext experiments not tied to particle physics. The consortium eventually set up at France's INRIA months later, with a third site at Japan's Keio University established by 1996.
Rough Consensus and Its Discontents
Berners-Lee's first instinct had been to work within the Internet Engineering Task Force (IETF), the volunteer organization that develops standards for Internet infrastructure. Founded in 1986, the IETF has no formal membership and no dues; everyone is a volunteer, and its standards are entirely voluntary and unenforceable. The organization operates on a principle of rough consensus, often confirmed by a show of hands or, at times, by asking members to hum for or against a proposal. It has been enough to sustain the organization for decades.
In March 1992, Berners-Lee flew to San Diego to propose IETF working groups for each of the web's primary technologies: HTTP, HTML, and the URI (later renamed URL). The process moved slowly. He needed a June meeting to formally propose the groups, and it took until the end of 1993 to persuade the IETF to set up all three. By then, new generations of browsers were shipping in months, not years, and the gap between specification and reality was widening.
The development of the web had also spiraled outside Berners-Lee's influence. Inline images, arguably the feature most responsible for the web's success, were invented during an late-night brainstorming session in a university lab basement. Berners-Lee learned about them when Marc Andreessen posted to the www-talk mailing list. He had hoped images might be treated differently, but the web was not his. As Andreessen later recalled, "Tim bawled me out in the summer of '93 for adding images to the thing."
Political HTML
Dave Raggett, a researcher at Hewlett-Packard, saw the web as commercial software. Unable to convince his bosses, he used the ten percent of time HP allowed for independent research to work on the web, becoming an active presence on the www-talk mailing list and at IETF meetings. In the fall of 1992, he visited Berners-Lee at CERN.
Around the same time, Raggett met Yuri Rubinsky, an advocate for Standard Generalized Markup Language (SGML), the language HTML was originally based on. Rubinsky believed HTML's limitations could be solved by stricter adherence to SGML. Raggett agreed, but was not ready to abandon HTML entirely. He began working on an enhanced version of the language and a browser to demo its capabilities, under the working title HTML+.
Raggett spent nights at a large computer that occupied a fair portion of his dining room table, surrounded by paper, crayons, and Legos. After a year of around-the-clock work, he had a version of HTML+ ready in November 1993, having added tables, images with captions and figures, and advanced forms.
Several months later, in May 1994, the first official web conference was held in Geneva, organized by CERN employee Robert Cailliau. Of the 800 people who clamored to attend, the space could hold only 350. Many were meeting for the first time, face-to-face after months on mailing lists.
The web stood on the precipice of ubiquity. Nobody from the Mosaic team attended, but rumors swirled about Marc Andreessen's new commercial browser. Mosaic had begun licensing its browser for commercial use, and an early Yahoo! was growing as major publications came online. Meanwhile, browsers were adding whatever HTML tags they wanted, forcing website creators to choose between cutting-edge technology and compatibility.
Dan Connolly, a young Texan working at the supercomputer manufacturer Convex, gave a talk called "Interoperability: Why Everyone Wins," describing a future where HTML fractured into competing implementations. His solution was an HTML standard able to evolve at the pace of browser development. The talk helped make Connolly chair of the IETF HTML Working Group.
Connolly and Raggett struck an agreement. Raggett would continue working on HTML+ while Connolly focused on a more narrow upgrade that would become HTML 2, an official standard that documented features browsers had supported for years. Raggett's proposal, renamed HTML 3, remained stuck, growing to 150 pages as it tried to accommodate an expanding web. "To get consensus on a draft 150 pages long and about which everyone wanted to voice an opinion was optimistic," Raggett later said. By then, he was already working at the W3C.
Deciding How to Decide
The W3C was officially announced in October 1994, with Berners-Lee joined by a handful of employees, including Raggett and Connolly. Any company or organization could join by paying a membership fee tied to company size. Member organizations would send representatives to provide input into the standards process. Despite the closed membership, the W3C operates in the open whenever possible: meeting notes and documentation are public, and experimental code is freely downloadable.
At the December meeting, the members settled on a process that stops just short of rough consensus. The W3C's technical specifications are known, in their final form, as recommendations — proposals that outline in detail how a technology works, but leave implementation to browsers. "We can't force members to implement things," said Sally Khudairi, former head of communications at the W3C. Drafts create a feedback loop between the W3C and its members; browsers begin to introduce specifications while they are still being drafted, and developers experiment with them, revising the draft until enough consensus has been reached.
By the end of 1995, the IETF HTML working group was replaced by a newly created W3C HTML Editorial Review Board. HTML 3.2, released entirely by the W3C and based largely on Raggett's HTML+, followed not long after.
Dynamic HTML and the Push for Standards
In 1997, browsers broke away from the still-new W3C recommendations. Microsoft and Netscape began shipping features under the banner of Dynamic HTML, or DHTML, and they almost split the web in two. Some features were added "without much regard for compatibility," as a 1997 Wired feature noted. Using DHTML required targeting one browser or another, and some developers simply chose a path, slapping "Best Viewed In..." banners on their sites. Others ignored the technology entirely.
Browsers had their reasons. Developers and users were asking for things not in the official HTML specification. As one Microsoft representative put it, "In order to drive new technologies into the standards bodies, you have to continue innovating." A more dynamic web was not a bad thing, but a splintered web was untenable. For some developers, it would prove to be the final straw.
The Three Working Groups
Following HTML 3.2, the HTML Editorial Review Board was divided into three parts. Dr. Lauren Wood, former theoretical nuclear physicist and Director of Product Technology at SoftQuad, chaired the Document Object Model Working Group. The DOM specification created a standardized way for browsers to implement DHTML. "You need a way to tie your data and your programs together," she said, "and the Document Object Model is that glue."
Chris Lilley, a computer graphics specialist from the University of Manchester who had worked on the HTML 2 spec at the IETF, chaired the Cascading Style Sheets Working Group. CSS was still a relative newcomer in 1997, with no major release yet. Lilley worked alongside CSS creators Håkon Lie and Bert Bos to create the first CSS standard.
The HTML group remained under Connolly, who had been around the web almost as long as Berners-Lee. He had been present in October 1991 when Berners-Lee demoed the web to an unimpressed audience at a hypertext conference in San Antonio. He contacted Berners-Lee a month later with just three words: "You need a DTD."
DTDs are what added the "S" (Standardized) to GML, the language IBM developed in the early 1970s to make it easier for typists to create formatted documents. Beginning with HTML 2, Connolly added a type definition to the language, limiting it to a smaller set of agreed-upon tags. In practice, browsers treated this loosely, but it was a first step.
In 1997, the HTML Working Group began work on the fourth iteration of HTML, which added more advanced features, complex tables and forms, better accessibility, and a more defined relationship with CSS. It also split HTML into three document type definitions. Frameset was rarely used. Transitional included the mistakes of the past, expanding HTML to accommodate non-standard, presentational tags like <font> and <center>, and was set as the browser default. The third, called Strict, pared HTML down to only its standard, non-presentational features, removing the unique tags introduced by Netscape and Microsoft. It drew a line in the sand: this is HTML. And it finally gave developers a way to code once for every browser.
The Web Standards Project
In August 1998, a small announcement appeared in Computerworld under the headline "Browser standards targeted," announcing a new grassroots organization of web developers called the Web Standards Project, or WaSP. Glenn Davis, co-creator of the project, was quoted: "The problem is, with each generation of the browser, the browser manufacturers diverge farther from standards support." At launch, 450 developers and designers had already signed up.
Davis was not new to the web. He had created Cool Site of the Day in 1994, featuring a single interesting homepage each day. As the site evolved into Project Cool, a resource for creating websites, Davis spent much of his writing explaining how to write code that worked in both Netscape Navigator and Internet Explorer. By the end of 1997, when both companies released their 4.0 browsers with spotty standards support and signaled that 5.0 versions would lean further into divergent DHTML extensions, Davis had reached a breaking point. He helped set up a mailing list with George Olsen and Jeffrey Zeldman. Within a few months, WaSP had tens of thousands of members.
"The W3C recommends standards. It cannot enforce them," Zeldman said, "and it certainly is not about to throw public tantrums over non-compliance. So we do that job." WaSP pushed browsers publicly and privately, using tactics like roadblocking: coordinating articles across several outlets at once where members worked, such as HotWired and CNet. Members were also asked to send emails to browsers and sign petitions showing widespread support for standards.
WaSP succeeded on several fronts. Opera had standards baked in at the beginning and was praised. For Netscape, after the browser was open-sourced as the Mozilla project, WaSP members noticed that its new Gecko rendering engine supported a fairly complete version of HTML 4 and CSS, unlike the planned Netscape 5 releases. WaSP pushed Netscape to include Gecko in its next major release, and the company agreed. Netscape 5 betas were standards-compliant but beset with other issues. By the time it shipped, Microsoft's dominance was near complete.
Microsoft was more insulated from grassroots pressure, but the company had an inside ally: Tantek Çelik, who ran the Internet Explorer for Mac team. Though it shared the IE branding, IE for Mac ran on a separate codebase, and Çelik's team was largely left to its own devices. He reached out to web designers in San Francisco, including members of WaSP, and asked what they wanted. "The answer: better standards support." That was all he needed to hear. Internet Explorer 5.0 for Mac, released in 2000, had across-the-board support for HTML, PNG images, and one of the most ambitious implementations of CSS yet seen. It would take years for the Windows version to catch up.
The XML Detour
Towards the end of the millennium, the W3C looked to a future where the web would exist on TVs, phones, and devices not yet dreamed up. The path forward, they decided, was eXtensible Markup Language, or XML.
XML traces its lineage to Yuri Rubinsky's advocacy. Rubinsky, who died suddenly in 1996, had devoted his career to making technology accessible, and he believed SGML — a meta-language providing a standardized way to describe the structure of data — was key. He had been having conversations with W3C members about bringing a more comprehensive version of SGML to the web. His influence brought together Dan Connolly, Lauren Wood, Jon Bosak, James Clark, Tim Bray, and others. In November 1996, XML was formally announced and adopted as a W3C Recommendation. It was essentially SGML, minus a few things, renamed and repackaged as a web language. It gave developers a way to define their own structured data with completely unique tags, and became the backbone of RSS, MathML, and server-level APIs.
XML's appeal extended to the maintainers of HTML, a language that "was beginning to feel somewhat complete," as Steve Pemberton, chair of the HTML working group at the time, later described. "Six months later, though, when XML was up and running, people came up with the idea that maybe there should be an XML version of HTML." The merger became XHTML, which within a year was the W3C's main focus.
The first iterations of XHTML, drafted in 1998, had stricter rules for authors to follow, opening up new possibilities. WaSP issued a press release lauding its capabilities. XHTML represented a web with deeper meaning, a Semantic Web that became the W3C's greatest ambition for close to a decade.
Subsequent versions introduced even stricter rules. XHTML 2.0, released in 2002, removed backwards compatibility with older versions of HTML, even as Microsoft's Internet Explorer — the leading browser by a wide margin — refused to support it. "XHTML 2 was a beautiful specification of philosophical purity that had absolutely no resemblance to the real world," said Bruce Lawson, an HTML evangelist for Opera at the time. Rather than uniting standards under a common banner, XHTML and the refusal of major browsers to fully implement it threatened to split the web apart permanently. It would take something bold to push web standards in a new direction — but that was still years away.
