The 2-writes + commit pattern helps with failed writes, but it still leaves race conditions open. Two concurrent requests can both succeed and you can still end up with inconsistent state across DBs without triggering rollback/error handling. The Canva talk’s recommendation was basically transactional outbox + small delayed reconciliation loops. Since the outbox write happens in the same transaction, you get better guarantees plus visibility into failed syncs/discrepancies. I think 2-writes is probably fine for simpler systems, but once concurrency/load increases you usually want some sync/reconciliation mechanism behind it. Worth watching the talk directly though, I’m paraphrasing and may be missing some details.