Automations you can read out loud

There are three ways the industry builds automation editors. A node canvas, where steps are boxes you wire together. A linear step list, where you pick an app, then an event, then map the fields. Or a sentence, where the rule reads as English with the variable parts underlined.

We chose the sentence, and the reason is narrow: we automate one product's own nouns. There is no app-picking step to make linear, and no arbitrary branching to lay out. A canvas would be ceremony around a rule that is one line long.

When, only if, then

A recipe reads in three lines. When a status column changes to Done, only if the record sits in a collection you name, then move it and tell someone. The blanks are inline and sized to their content — dashed while empty, solid once chosen — so an unfinished rule looks unfinished rather than looking broken.

We built a plain-English preview underneath the form early on, then deleted it. If the form itself reads correctly, a second block restating it says the same words twice — and wanting one at all was the proof the form was not yet readable.

Offering only what can run

• A trigger that fires on a sub-record never offers actions that only make sense on a parent — the engine would refuse them, so the list does not show them.
• A workspace-wide rule hides the collection picker, because naming a collection would quietly narrow the rule back to a single module.
• Impossible options are filtered out, never greyed out. A disabled option still leaves people wondering what they did wrong.

Every run lands in the activity log, stamped with the recipe that caused it and the person whose edit set it off. There is no separate run feed to go stale — an automated change is an activity row, so that is where it lives.