History of the Internet

Google Buzz Turned Gmail Contacts into Followers. “No” Wasn’t a Clean Exit.

Google pitched Buzz as a social network with the setup already done: the people a Gmail user emailed and chatted with most would become the first people they followed. The shortcut moved a relationship built in private correspondence toward a public social product. The Federal Trade Commission later described what happened when users declined the invitation or switched Buzz off: some could still be followed or appear in other users’ social displays. Google accepted a settlement requiring a new consent rule and privacy checks that would outlive Buzz itself.

Mozilla Renamed Its Browser Firebird. Firebird Was Already Here.

Mozilla's small browser began as Phoenix. A company using that name objected, so Mozilla chose Firebird. Then the Firebird database project objected too. By February 2004 the browser was Firefox. The sequence sounds like a branding department losing three rounds of a word game. The record is sharper: one open-source project had to decide whether its own legal comfort outweighed another project's established name and the distributors both relied on.

Apache Refused Sender ID. The IETF Put Its Rival on Trial Too.

In 2004, the Apache Software Foundation told an IETF mail working group that its SpamAssassin and JAMES projects could not ship Sender ID under Microsoft's patent-license terms. Two years later, the IETF published Sender ID beside its rival SPF as simultaneous experiments, warning that a receiver might read the same DNS record through the wrong test and lose legitimate mail. The standards body could publish both designs. It could not make implementers accept a license or make operators deploy a protocol.

Google Put Comments Beside Any Webpage. Then It Closed the Sidebar.

A website owner could publish a page and leave comments switched off. In September 2009, a visitor with Google Toolbar could still open a separate sidebar and write beside that page. Google called it Sidewiki. Google chose which entries rose to the top and could carry an annotation to another page containing the same passage. Two years later it decided to discontinue the service and offered contributors time to download their words. The page remained where its owner put it. The conversation beside it had a different landlord.

Wikipedia Had One Summer to Change Its License

Wikipedia's authors were scattered across years, languages and usernames. By 2008, their encyclopedia had a licensing problem no editor could fix by clicking Save: its GNU Free Documentation License did not work neatly with the Creative Commons license used by other free-content projects. The Free Software Foundation offered a narrow door. It let an eligible mass collaboration's site operator republish the work under CC BY-SA, but only before August 2009. Wikimedia had to ask its contributors, make a board decision and change the site's terms while that door was open.

Library Took the Tweets’ Text. The Pictures Stayed Outside.

The Library Took the Tweets’ Text. The Pictures Stayed Outside.

In 2010, the Library of Congress said it would save the stream of public tweets. The small print mattered: it would receive text, not the pictures, video or linked pages that often carried the rest of the post. Seven years later the Library pointed to that gap while ending its attempt to collect every public tweet. It kept the twelve years already acquired and narrowed future intake. The archive had not lost its files. The meaning of a tweet had grown beyond the part it was taking.

Host That Never Took Mail Kept Everyone's Mail Waiting

The Host That Never Took Mail Kept Everyone's Mail Waiting

In 1995, a machine devoted to DNS or time service might never accept email. A person could still send a message to an address at that machine. If no mail program answered, the sender's system had to treat the failure as temporary and try again. Days could pass before a confusing bounce reached the person who had made the mistake. Two engineers proposed an odd cure: a computer that did not run a mail service should sometimes run just enough of one to say, decisively, that it did not run a mail service.

Google Sold the Download. Then It Told Buyers the Video Would Stop Playing.

In 2006, Google offered a video store with files people could buy and download. In 2007, it told those customers that purchases already on their computers would cease to play. The company reversed course after the objection landed: a card refund replaced store credit, and playback got six more months. The file had travelled home. Its useful life still ran through the seller.

Flickr Kept Your Photo Address. Yahoo Took the Key to the Door.

In 2005, Flickr told its photographers they could keep their screen names and public photo addresses. The thing changing was the key: new users would sign in with a Yahoo ID, while old Flickr users could keep their email-and-password login for a while. By 2007 that exception was closing. A photograph might remain at the address everyone knew, yet the person who made it now had to pass through a different company's account system to manage it. Public identity stayed put; entry changed hands.

SmugMug Bought Flickr. Yahoo Still Held the Login.

The photographs kept their Flickr addresses when the service changed hands in 2018. Members still entered with Yahoo credentials. It took a separate login migration in 2019 to begin handing them a new key. Flickr's own notices show how a public archive, a corporate owner and the private route into an account can move on three different schedules.

XO Laptops Could Talk. The School Server Still Had to Buy the Internet.

The XO Laptops Could Talk. The School Server Still Had to Buy the Internet.

One Laptop per Child put a radio in every child's machine and invited the machines to relay for one another. A classroom could become a network before a telecom carrier supplied a connection. Then the network reached the school gate. Internet access, shared storage, address management and updates accumulated around a separate server and a separately financed uplink. The hardware story was one laptop per child; the operating story had another computer in the room.

AOL Sold Bebo. The Advertisers Were Still in Its Accounts.

AOL bought Bebo as a way to gather users and sell advertising. Two years later it disposed of almost all the social network's assets. The filing for that year still expected money from Bebo's old ad customers on AOL's own sites—and kept Bebo in continuing operations. The website, the tax deduction and the sales relationship did not leave through the same door.

Red Hat Bought Two Netscape Security Systems. One Opened First.

Red Hat announced a deal for Netscape's Directory Server and Certificate Management System in September 2004. The purchase closed in December. By June 2005, Directory Server code was open under the GPL and the supported product was offered through an annual subscription, with purchases projected for mid-June; Red Hat was still promising to release more of the acquired technology. The contract had already split software, patents and trademarks into separate parcels. The word “acquired” concealed a long line of decisions about what buyers, developers and customers could actually use.

Website Moved Its Filing Cabinet. It Sent the Moving Bill to Every Link.

The Website Moved Its Filing Cabinet. It Sent the Moving Bill to Every Link.

A publisher renovates its site. The document is still there, the staff still believe in it, and the new menu looks splendid. The old address now leads nowhere. In a 1998 essay, Tim Berners-Lee treated this familiar renovation as a choice about who does the work. A site can move files, replace software and change departments behind a stable public URI. When it changes the URI instead, every reader, bookmark and distant publisher must rediscover the document. The office got its fresh floor plan; the rest of the Web inherited its missing address.

WebCrawler Had a Robot for Each Query. Pinkerton Kept It Behind the Gate.

The public WebCrawler in 1994 answered questions from a shared index. Brian Pinkerton had built another mode: start with likely results, send a robot down their unexplored links and add whatever it found before answering. That sounds like the perfect cure for a search index with holes. Pinkerton kept it out of public hands. A shared crawl could serve thousands of queries; a fresh crawl for every curious user would ask the young Web to pay the network bill again and again.

OpenStreetMap Shut Off Edits to Give Changes a Name.

In April 2009, OpenStreetMap contributors could not save new edits through the project API while the service moved to version 0.6. The online editor was disabled. When editing returned, every change had to belong to a changeset, and clients had to check which version of a map element they were replacing. A community map paid for a better account of its own making with a conspicuous pause in the act of making it.

Webmention Delivered the Reply. The Other Site Kept the Door.

A blog author could write a reply on her own site and still want it seen beside the article she answered. The old choices made her pick a house: put the comment on someone else's page, or keep the reply at home and hope readers followed the link. Webmention gave the link a messenger. In its 2017 W3C Recommendation, the sending site tells the receiving site where the reply lives. The receiver checks the link and decides what appears on its page. The messenger crossed the street; the front door stayed with the owner.

Site Was Down. Google Had a Copy. Then the Link Vanished.

The Site Was Down. Google Had a Copy. Then the Link Vanished.

For years, a Google result could offer two doors to one page: the publisher's site and a snapshot stored by Google. If the site failed, the snapshot might still open. The publisher could ask Google to remove that copy. In 2024 Google removed the reader-facing cached link itself. A little secondary doorway had exposed a long argument about who controlled a page after it left its own server.

Internet Explorer Made the Reader Click. The Patent Case Was Offscreen.

In 2006, some embedded content in Internet Explorer appeared on the page but waited for a manual activation before a reader could use it. Microsoft connected that change to its Eolas patent case. A legal and engineering decision had acquired a user interface: click, or use the keyboard, and only then interact. Two years later another Microsoft update removed the extra step.

Creative Commons Made Licenses Machine-Readable. Then It Hid the Label in a Comment.

On 16 December 2002, Creative Commons launched licenses that could speak to three audiences: a human reader, a lawyer and a machine. The machine was supposed to find works offered for reuse without making every person negotiate permission from scratch. The first recommended web markup carried RDF/XML inside an HTML comment. Six years later, Creative Commons' own ccREL paper called that choice cumbersome and replaced the recommendation. The legal permission was only one part of the work. A machine also had to find the label attached to the right object.