LLMs and LLM agents are very powerful at finding relevant information, provided you are able to ask what you are looking for.
Ontology work (building taxonomies, etc.) is especially useful now, since it enables you to ask the right questions, and frame your research in the right way.
A corollary is that Domain driven design is more important and accessible than ever. You can stress-test a design against agent-sourced research.
Note that neither of these use cases actually requires you to maintain or rely upon LLM-generated text. Research can provide you useful primary source documents (which can be curated into subsequent model context), and stress-testing a design does not mean trying to automate the design process; it just helps de-risk a project.