DuckDB works well if

* you have a small datasets (total, not just what a single user is scanning)

* no real-time updates, just a static dataset that you can analyze at leisure

* only few users and only one doing any writes

* several seconds is an OK response time, get's worse if you have to load your scanned segment into DuckDB node.

* generally read-only workloads

So yeah, not convinced we lost a decade.