just dont fall in love with goofy memory style features and its likely gonna be fairly easy to just plug and play whatever model for a ton of use cases.

I dont see a lot of love for weird memory like features on HN, but on provider subreddits its constantly talked about.

Every project I have tends toward coding agents over time partly for this reason - if nothing else, I want have both control, visibility and portability over memory and a log of the reasoning that went into the current state of things.