RAG is a type of information retrieval process. It modifies interactions with a large language model (LLM) so that it responds to queries with reference to a specified set of documents, using it in preference to information drawn from its own vast, static training data.
Will it still be used by cutting-edge technology companies in 2026 (EOY)?
Resolves NO if prominent players, e.g., Microsoft, Google, perplexity (or new entrants) publicly pivot away from RAG as mean of generating responses.
Resolve YES if this is still the state-of-the-art.
All forms of RAG count towards the resolution.
Agentic AIs that can look up sources on their own do not count.
Agreed that it is vague.
IMO, there is a qualitative difference between
1. A series of functions doing retrieval, dumping all into a context, and then running LLM for a generation.
2. An agent running on a cluster, that decides to browse the web or a DB and then outputs results using this information and its training.
The former is RAG, the latter is closer to AGI.
Perplexity pro is a complex series of functions, but still very much in the shape of RAG.