The problem
Dynamic HTML let a page change itself, but the data still lived on the server, and reaching it meant a request that replaced the whole page: the screen went blank, the scroll position was reset and the user waited. Next to desktop programs, web apps felt slow.
The idea
A script sends its own HTTP request with XMLHttpRequest and carries on; when the answer arrives, a callback updates just the part of the page that changed. Microsoft shipped the object as XMLHTTP in Internet Explorer 5 (1999); in 2005 Jesse James Garrett named the approach Ajax, "shorthand for Asynchronous JavaScript + XML".
Here the page loads once, without the list. request() opens an XMLHttpRequest, sends it and waits in onreadystatechange until readyState is 4. The server answers every request with the whole list as JSON, and render() rebuilds only #todo-list, the counter and the filter buttons. Filtering happens in the page, without a request.
What hurt next
Every browser did it a little differently. Internet Explorer 5 and 6 created the object with new ActiveXObject("Microsoft.XMLHTTP"), Internet Explorer before version 9 attached events with attachEvent instead of addEventListener, and each browser had its own DOM quirks, so scripts grew long and full of browser checks. Libraries like jQuery wrapped XHR, events and the DOM in one short API that worked the same in every browser.
How close this is to the original
The demo calls new XMLHttpRequest(), the native object of Internet Explorer 7 and the other browsers. Internet Explorer 5 and 6 only had new ActiveXObject("Microsoft.XMLHTTP"), so real code checked window.XMLHttpRequest first and fell back to it; ActiveXObject belongs to Internet Explorer and does not exist in Chromium. The "X" was XML: early code read responseXML, or turned JSON text into objects with eval(), because a built-in JSON.parse arrived only in 2009 (Internet Explorer 8, Firefox 3.5) and was standardised in ECMAScript 5 that December. The demo answers in JSON and reads it with JSON.parse.