Sharding the Prototype

A product has 400 users and one Postgres instance at 3 percent CPU. An engineer proposes sharding the database by tenant now, arguing it will be painful to retrofit later. What is the strongest argument against doing it in this sprint?

A single Postgres instance can be scaled vertically without limit, so a sharded topology is unnecessary at every stage of growth

Sharding never helps write-heavy workloads, so the effort would produce no capacity benefit at any user count the product reaches

Premature optimization means performance work should be deferred until users complain, so no capacity work belongs on a roadmap at all

It locks in a partition key before the real query patterns are known, and a wrong key is harder to undo than adding sharding later