Early-stage teams are usually told to move fast and not overthink infrastructure — and for most decisions, that's correct. Framework choice, hosting provider, even initial database engine are all more reversible than they feel at the time.
A smaller set of decisions are genuinely expensive to reverse: the core data model for the primary entity in the product, the authentication and multi-tenancy approach, and anything touching how customer data is partitioned. Getting those wrong doesn't show up as a problem until the business has scaled enough that fixing it means a migration project, not a refactor.
The practical rule: move fast everywhere except the handful of decisions that are hard to undo, and spend deliberate time on exactly those.