It didn't turned by default (as of 21.11)
It waits 200ms by default to collect all concurrent queries from all of the clients than group all this records together to insert at once.
The clients will get the answers only after this batch will be flushed to the target table. Only after that this data will be available for SELECT queries.

From 24.2. they 200ms is adjustable automatic.

This memory buffer is limited 200ms and by volume 100k records.

The data in this batch will be reorder and deduplicated if you insert in ReplacingMergeTree.

It works on all table types including distributed. It can work even with mySql table.

It replaces the Buffer Table engine. You don't need it anymore. It's essential the same.

optimize for inserts

asyn_inserts enable
wait_for_async_insert enable