Pick a specific area, read lots of papers, find out what the bleeding edge statements 'we can do X' and 'Y would be nice and seems within reach but we're not doing it now' are there.
Do something interesting and novel based on that - not neccessarily a detailed implementation, but at least a quick&dirty proof of concept; and then also describe and publish it as a formal paper, it's not particularly much work compared to actually doing the stuff. For example, if you "want to work on tools that advance how we think about software" then picking some specific narrow area of how, say, Bret Victor's concepts could/should be implemented in real practice would be an interesting topic.
All of this can be done w/o a PhD as a hobby/side-project, and such a project+publication would also be a ticket to a decent PhD spot.
Do something interesting and novel based on that - not neccessarily a detailed implementation, but at least a quick&dirty proof of concept; and then also describe and publish it as a formal paper, it's not particularly much work compared to actually doing the stuff. For example, if you "want to work on tools that advance how we think about software" then picking some specific narrow area of how, say, Bret Victor's concepts could/should be implemented in real practice would be an interesting topic.
All of this can be done w/o a PhD as a hobby/side-project, and such a project+publication would also be a ticket to a decent PhD spot.