In most enterprise projects, the part that breaks most often isn't the Salesforce org itself - it's the stitching between it and everything else: ERPs, legacy systems, messaging queues, other SaaS products. Integration is where architecture actually gets tested, because failures show up under volume and partial failure, not in a clean dev environment.

That means deliberately choosing between synchronous and asynchronous, designing for idempotency (the same event processed twice can't create two records), and treating integration failure as an expected scenario, not an exception - with retries, dead-letter handling, and error visibility, not just a silent try/catch.

Platform Events and replay patterns deserve special attention: they're powerful, but easy to misuse when treated as a generic Pub/Sub without thinking through ordering and replay windows.

The Platform Integration Architect certification reflects exactly this area - integration decisions are architecture decisions, not an implementation detail to figure out later.