A larger context window does not remove retrieval

Cost grows linearly

Sending the whole corpus on every call means paying users multiplied by document volume. Retrieval turns that product into a handful of relevant chunks.

Freshness is still your problem

When a document changes you still need to know what to re-send. That decision is indexing. Stuffing everything just defers it to every call.

You cannot cite a source

If you cannot answer "which document did this come from", a human cannot verify it, and what cannot be verified cannot be used where mistakes are irreversible.

So what changed

Bigger context means retrieval can afford to be looser. Where you once had to pick the top 3, you can now include 12 — trading a little precision for recall. Retrieval did not disappear; it stopped needing to be so accurate.