A copy of a database increases read capacity, but the copy may be behind. You must decide which reads can tolerate that delay. Elm Workshop can show an old public profile for two seconds; after a person changes their own address, returning the old address makes the save look broken.
Separate durability from freshness. An asynchronous replica can lose acknowledged writes during a primary failure. A synchronous acknowledgement waits for another copy and adds latency. Neither choice removes the need to define failover and prevent two primaries from accepting conflicting writes. Use the workload to choose a read route and an acknowledgement rule, then describe what a user sees during lag or failure.