Watch the work first
Before automating a task, follow it through a few real repetitions. Notice which steps use a stable rule and which depend on context. A repeated task is not always a predictable task. Write down the exceptions before choosing the tool.
Design for the second attempt
Workflows should have a clear state and a safe recovery path. If a service times out, can the action be retried without creating a duplicate? If a step needs review, can someone see what happened and continue from there? Reliable recovery is often more valuable than a perfectly smooth demonstration.
Leave judgment visible
Make the automated steps understandable. Keep a useful history, surface failures, and provide a way to pause the workflow. The goal is to reduce repetitive effort while keeping people able to make the decisions that matter.