Most rewrites are not caused by bad code. They are caused by decisions that assumed today's traffic, today's team and today's product would stay roughly the same.
We design the seams early: clear service boundaries, boring data stores, and a path to split a module without stopping the business. That does not mean over-building. It means refusing accidental architecture.
A practical rule: if a part of the system might grow 10x, give it a contract, observability and an owner. Everything else can stay simple until the data says otherwise.