I never used i686 even though I built for it.
The way I built packages was on my local laptop (clean chroot). Building for 2 separate architectures involves building 2 separate packages. This doubles the time it takes for a developer to build. This might not be that bad for smaller packages, but ghc (haskell compiler) took a long time on my computer. This means that if it takes an hour to build ghc, it would take 2 hours of my time to actually build ghc for both platforms. When you're talking about 100s of packages built many times over several years, this quickly adds up.
The other benefit from removing i686 support is that the packages built for that architecture were never really tested by developers. I ran x86_64 10 years ago, tested the x86_64 version of my packages, and most of the time pushed them directly to the main repositories if everything checked out. For more critical packages, I would push them to testing, let people report any issues, and then move them after a couple of days. This means that I never knew if i686 broke or not. This was the case with most of the developers at the time.