Most RAG implementations still use fixed-size chunks. This approach dynamically adjusts context boundaries based on the user's query intent, which solves a lot of the "context clipping" issues we see in production.