For the most part it didn't matter until recently.

Netscape versus IE (late 90's) - features didn't matter, they just rendered HTML and CSS differently

Firefox versus IE (early 00's) - Firefox added a bunch of great CSS support and things like rounded corners and PNG transparency, so once we could use those we could just supply a polyfill for the specific IE version that needed it. Opera could handle it already, Netscape was Firefox rebranded. Only IE (which announced itself as IE) needed extra help

Mobile vs Desktop (late 00's) iPhone! Android! Tablets! Now is where things start to get a little crazy, IE will be IE but a bigger concern is the separation between tiny little touchscreen devices browsing a website, and a massive desktop computer

Mobile versus Mobile (early 10's) - IE never says it's IE, we have smart watches, phones, phablets, tablets, netbooks, notebooks, and still desktops. There is Firefox, and Chrome that run on Mac, Windows, Linux, iOs, and Android, there's Safari which runs on OS X, Windows (old version), and iOS, there's IE, of which there are 8,9,10,11 and the new ones in circulation, and a handful of other browsers like Android browser that kind of gave up a long time ago but are still used.

I'm sure backend software was rife with feature-detection for the OS's it ran on (Redhat versus CentOS, special support for IIS, etc) but until things exploded after the iPhone was released in 2007 the web had very predictable deficiencies that could be addressed more directly than feature detection.

I can remember as a Linux user, there were plenty of websites that would only let 'approved' User-Agents in, because they would rather you NOT see their site than see a site in a browser they didn't support. When using Linux I often didn't have access to IE or Netscape, so I would use Firefox or Konqueuror to spoof a different User-Agent. Nintendo.com used to be like this, plus others.

1) it didn't take "web folks" long. It was near immediately after the availability of the browser as an application platform that the advice took hold to those willing to heed it.

2) The existence of autoconf et al is not in any way preventing people from performing platform tests all over the place, so it's not like a panacea.

Sorry if you find this a bit abrasively worded, I aimed to avoid that, but you made me bristly with your implication that "web folk" are somehow lesser.

Autotools don't only check for features, it also checks for bugs. For example I know that Firefox SVG support is buggy and I need to use Canvas version of OpenLayers for that browser - how am I supposed to do it without UA sniffing?