I do a lot of urban architecture shots with a DX camera, civil engineering (bridges, building sites, etc.), and some landscapes. My first foray into wide angle was 14.5mm, which I found annoyingly restrictive. I graduated to a Sigma 10-20mm, which I liked a lot, but checking over my shots found that a high proportion of them were taken at 10mm. In other words I had very often rammed it as wide as it would go. That confirmed my subjective impression that something wider would be really useful. I got a Sigma 8-16mm. Those extra 2mm were really important. They made the big difference between struggling with 10mm to just reach an impression of the existence of all four walls of a quadrangle while shooting from inside it, to being able to easily get all four walls in with a bit of extra latitude for geometry correction and perspective adjustment.
The number of shots I take with the 8-16mm rammed up against 8mm and wanting a bit more was very much less than with the 10-20mm. But that phenomenon was still there. Sometimes, although much more rarely, I still wanted wider. But 8mm on a crop sensor is close to the practical limits of what can be achieved with a rectilinear perspective projection lens. As far as I know there's nothing wider. For those rare occasions when I want to go wider with a rectilinear view I use a fisheye lens, and use a combination of partial defishing and cropping to get a near enough approximation to linearity. That kind of post processing loses a lot of resolution. Where that matters using a tripod and stitching together perspective adjusted panels is the rather difficult and tedious solution.
Sometimes, such as with panorama landscape shots, what is wanted is a narrow but wide shot with a cylindrical perspective, i.e. linear in the vertical dimension and fisheye (preserving subtended angles rather than linearity) in the horizontal. That is best accomplished by stitching a panorama together. Some of the more modern camera bodies now offer a surprisingly good ability to do handheld panorama sweeps which the camera stitches together into a panorama jpeg. It's tricky to spin slowly and smoothly around without wobbling a bit in speed and aim. The camera panorama stitching process gets rid of that by losing resolution. Much better results can be got is you use a tripod with a fluid (video) panning motion. That makes it much easier to keep the aim and speed of pan straight, improving resolution. For ultimate resolution of panoramas of course you must take the individual shots yourself and process and stitch them together afterwards.
All this is irrelevant if you don't have a hunger for extreme wide angle and regard the various perceptual and geometric distortion problems as annoyances to be avoided rather than interesting obstacles to be overcome
