How do I figure out what to actually build instead of just taking requests?

You take feature requests well, but figuring out what to actually build is a different job, and it makes sense that it feels out of reach. Taking requests is reacting. Discovery is deciding, and deciding means sitting with uncertainty until a real problem comes clear. That is harder and less comfortable. A list of requests is not a direction. It is a pile of symptoms. If you build each one as it comes, you get a crowded product that solves nothing deeply. The thing worth building usually hides underneath the requests, in the problem they all circle without naming. Take your last ten requests and look for the pattern under them. What are people actually trying to get done? Name that one job. Then ask whether any single thing would serve it better than the ten features would. Discovery is not doing more. It is looking for the one problem hiding behind the many asks.