The post focuses on the least safe mode (NONE), and barely touches on REPLICA_SAFE at the end of the article. Leaving the reader to wait for "a future blog post".
Choose the tool and configuration for the job. For a Mongo user that needs fault tolerance and consistency guarantees, I would recommend (based on documentation) choosing a configuration that attempts to provide those.
The safest server configuration would be a multi-node replica set (3 or 5) that spans multiple data centers (preferably 3 or more). The safest client configuration would be a write concern REPLICA_SET (majority).
If you have examples of Mongo not providing fault tolerance in this setup, the world would be a better place if you shared.
The post contains the 'straw man' informal fallacy - ie. standing up an absurd scenario in order to knock it down.
Mongo provides write concerns for several use cases. Those concerns are:
NONE
NORMAL
SAFE
FSYNC_SAFE
REPLICAS_SAFE
The author focuses on the write concern NONE - a write concern that is provided for the 'fire and forget' use case. To choose NONE and expect guarantees and fault tolerance is simply user error.