Any specialty with a decent amount of scut work involved on a day to day basis (ie work that doesn't involve real high-level medical knowledge or decision-making) will have a role for midlevels, since they usually aren't any worse at doing scut than physicians. Specialties like EM and hospitalist can be suspectable since they are not as highly specialized and their work is often done in consultation with other specialties.
In surgical specialties, you can't expect a PA or NP do perform the whole surgery themselves, but surgeons in their groups frequently hire them as extenders to do stuff like write notes for consults or round on post-op patients on the floor.
While studies have shown the downsides of using unsupervised midlevels in inpatient settings (eg higher resource utilization, more reliance on consults, longer length of stay) these problems can be mitigated by using only experienced midlevels. Some hospitals, for example, consider a hospitalist midlevel with 5+ years of experience as comparable to a junior attending that just graduated from residency. Hence experienced midlevels are very valuable since they are probably paid half as much or less per hour than a junior attending.
In the outpatient setting, midlevels may have more limited scope, since in non-emergent care patients have a choice of provider. Any many, especially those with means and better insurances, will specifically request to see only a physician and not a midlevel. Midlevels may therefore be providing more care to lower-income patients with poor insurance who probably don't have as much of a choice.
Agree that radiology and path are difficult for midlevels to be useful since there is minimal scut and every study requires high-level expertise and there is little room for error.