If you can run your system on a single machine, you don't need an SRE.

If you have hundreds or thousands of machines, that's an indicator that you /may/ have the complexity that requires the disciplines that can come from dedicated SRE. The tough thing is conflating filling operations problems with a role named SRE, versus actually using the best practices that will help you scale and improve reliability.