A recorder repeats exactly what happened; it does not understand why the action was correct.
Recognize poor candidates
Avoid it when the screen changes frequently, the task requires judgment, errors are costly, data is sensitive, or a native API or automation action already solves the problem.
Use the starting-state test
MacroLoop is most effective for stable, low-risk repetition. If you cannot define one consistent starting screen and one safe ending state, do not create an unattended loop.
Keep high-impact work supervised
Money movement, legal acceptance, permissions, deletion, public communication, and customer-data decisions should not depend on coordinates alone.
THE TAKEAWAY
The clearer the exceptions and consequences, the less suitable blind playback becomes.OFFICIAL REFERENCES