Automation that survives production
The small operational decisions that turn an n8n workflow from a demo into a system people can rely on.
Make failure visible
Every external call can fail. Capture the input, response, and retry state so a broken run is diagnosable without reproducing it locally.
Keep steps idempotent
Retries should not create duplicate posts, records, or notifications. Use stable operation keys and check downstream state before writing again.
Design the handoff
Automation still needs a clear owner. Define which failures retry automatically, which require review, and what information the reviewer receives.