agents subsume embeddings
agentic RAG can use embeddings-based retrieval as one search tool among many. it remains to be seen whether embeddings will be able to justify the additional complexity and costs.
embeddings: legal vs medical vs software, each domain needs different types of search reasoning/strategy, dependent on how each domain uses the information at its disposal.
dark horse: graphRAG is promising. we may get models that specialize in converting text to knowledge graphs. i've seen a few cool releases from microsoft on this front.
hunch: advanced graphRAG will give ai superpowers. thinking like it's never been done before, with perfect memory. humans do graphRAG, but we can't do advanced kg ops. imagine an LLM with 17 different rigorous ways to find related concepts and factoids.