The metadata model can be changed (within reason) after being deployed. Things you should not do are:
- removing / renaming existing types / aspects
- removing / renaming / moving existing properties / associations
- changing value types and index configuration for properties
- changing the referenced class for associations
- changing multiplicity / mandatory-ness of properties / associations
- adding new mandatory properties / associations to existing types / aspects
- adding new mandatory / default aspects to existing types / aspects
- changing type / aspect inheritance
You can technically still do all these things, but then you will have to write some extra code (migration / adaptation logic for existing data) and perform some administrative tasks (re-indexing SOLR) to make that work properly.
I have worked in multiple projects using various variants of "agile" methodoligy, and model evolution has so far never been an issue / hindrance. Of course it helps if you are already familiar with the above mentioned restrictions, mitigations and have some experience in how some use cases may evolve from previous content management processes to better inform your content modelling plans / techniques.