AI Search review

AI Search (formerly AutoRAG) packages a retrieval-augmented generation pipeline as a managed service. Point it at an R2 bucket or website and it handles chunking, embedding, indexing, and querying. Most teams have better things to do than build their own RAG pipeline, so the idea makes sense.

It still feels early. When results are bad, the indexing pipeline gives you little help understanding why it retrieved the wrong chunk. Configuration is limited, and the default Workers AI models carry the reliability concerns I cover in that review, though you can now route generation through other providers via AI Gateway.

For an internal docs-search bot where mediocre answers are acceptable, it’s a fast win. For anything customer-facing, I’d wait a few more quarters and watch the changelog.