The convenience of serverless data systems is real. It is also easy to misunderstand. Removing cluster management does not remove systems thinking.
Athena and Glue workloads still care about partition design, schema evolution, file size, table metadata, compression, and how users actually query the data. Bad layout can turn simple questions into slow and expensive scans.
The discipline is to treat convenience as an interface improvement, not an architecture exemption. The system still has physics. It just moved where the operator has to look.