I understand them in terms of a map-only form of container-scoped agent: the smart adornment can’t match items outside the current map. However, it is another agent in the agent cycle (albeit testing a smaller maximum number of notes).
Any query/action being evaluated has to use some resource. But, I think you are phrasing the question incorrectly. Try:
“Should I worry about using smart adornments?”
No. If you find your document is sluggish compared to others, large amounts of ‘agent’ queries may be the problem. But I’d not worry about a problem you don’t have. But, do review your work pattern. Are you using a lot of queries/code for the convenience of not remembering anything? Do remove or turn off, or throttle back un-needed or unimportant queries.
If every new document task adds lots of queries, look at more efficient query structure, e.g. (with no more detail I can’t put that more precisely). Are 10 different things all polling the same large set of data? If so, consider using 1 to do the big initial search an others within it. Think about how scale (more data, tasks, etc.) acts on your document.
There is no simple one-size-fits-all answer. But, don’t worry unduly about slowdown if you’re not encountering it. Nothing is more corrosive to progress than no acting in case some unspecified bad result occurs. Deal with the issue when it occurs.