Ismael Celis's point: durable execution normally requires snapshotting each step of a workflow so that on failure you can resume from the last successful step. In an event-sourced system, each persisted event already records what the last successful step was. If a reaction fails (e.g. Stripe down, bug in your code), you fix it, redeploy, and the runtime retries just that reaction from its last good event — no re-running of earlier side effects.