Tim Berners-Lee's 1998 Case for URIs That Outlive Their Servers

This W3C essay argues that a well-designed web address should still resolve decades after it was created, and that broken links are almost always a failure of organizational discipline rather than technology. Nothing about HTTP forces a URI to be unstable — servers like Apache let you decouple the public address from wherever the underlying file actually sits on disk, so the excuses people offer (redesigns, staff turnover, moving to scripts) are really symptoms of putting implementation details into the address in the first place.

The practical advice is to bake as little changeable information into a URI as possible. Author names, directory layouts, and mechanism markers like ‘cgi-bin’ or ‘.pl’ all leak how the site works today and guarantee breakage when that changes. Creation dates are the notable exception: a year embedded in the path (as in the NSF’s ‘/pubs/1998/’ example) signals a versioning scheme and ages gracefully. It also pushes back hard on the idea that URN research will someday rescue sloppy link management — if your organization can mint persistent identifiers, prove it by using them as your HTTP URIs now, backed by a database that maps stable names to current file locations.

The stakes are reputational and behavioral: every dead link erodes trust in the site owner and frustrates users trying to get something done. Treating URI assignment as a deliberate design responsibility — one a webmaster should stand behind in two, twenty, or two hundred years — remains the essay’s core, and still-relevant, demand. Berners-Lee concedes the tooling of the era made this genuinely hard, framing several points as ‘do what I say, not what I do.’