(1) You can tell people to go to a URL and it almost always "just works". Contrast that to considerable work to develop a Windows installer, Mac DMG file, RPM/DEB/other packages for Linux, etc.
(2) Once you put installers and packages in the field you will discover some people have machines in an odd state and your software won't work on them. Some people will give up (either your competitor wins or indifference wins), others will lead you into a rabbit hole solving their problems.
(3) Something will go wrong with your software. For a low risk change, say fixing a typo, you can deploy a change to a web app in minutes even with the discipline of version control, automated tests, and cache invalidation in place. For other platforms the process of building and testing installers, wait for/fight with the app store, then getting the people to really upgrade is a huge hassle that could stretch on for days or weeks.
(4) If you don't have the software versions synchronized, client/server protocol mismatch will complicate your efforts to fix bugs and add features.
(5) Data is safer in a data center or a cloud than it is for a client device that could can be lost, stolen or destroyed in a opportunistic or targeted way
(6) No cross-platform UI framework is better than the web browser in terms of (a) UI quality (b) ease of development or (c) range of devices supported.
(7) Any system for e-commerce, internet of things, digital twins, machine learning, space travel, compilers, data structures, weapons systems, etc. has a user interface on the critical path
My heroes right now are people like Walt Disney and Jim Henson who built seductive systems and spaces to enchant people. I find that opportunity meaningful in U.I. work.