The best candidate for automation is not always the task that takes the longest. Sometimes it is the small interruption that repeatedly breaks your attention.
Renaming files. Collecting a few exports. Checking whether a report arrived. These tasks are individually minor, but their switching cost can become the shape of an entire morning.
Write down the rule first
If I cannot explain the task without saying “you just know,” I probably do not understand its boundary well enough to automate it completely.
A useful description separates mechanical work from interpretation. “Collect files from this folder and group them by date” is a rule. “Decide which files matter” may involve context that is not captured anywhere.
Automate the first part. Make the second part easier to review.
Make mistakes reversible
For a file-organizing script, the first useful output is a proposed list of changes. Show the original path and the intended destination. Detect collisions. Save a record. Only then consider moving files.
The same pattern applies to more complicated workflows:
- Preview the action before committing it.
- Keep enough information to trace what changed.
- Separate retrying a task from repeating its side effects.
- Stop when the input falls outside the rule.
Keep the human decision visible
A good automation removes busywork without quietly taking over decisions nobody meant to delegate. That distinction is especially useful when adding AI to an existing process.
Start with a small, frequent irritation. Write down the rule. Build a preview. Use it long enough to learn where the rule was incomplete. The goal is a little more attention available for the work that actually needs it.