Hacker News Reader

Comment by AlexErrant | Hacker News Reader

> If you can't generate a report, it may be because you don’t have Memory turned on.

Has anyone found success with memory and Claude Code? I found Opus's 4.8 memories largely lacking value. Memories are too verbose/specific/yet-generic for the value captured.

Instead, I've been holding "retros" with the agent immediately after a session, and _those_ responses have been typically fantastic. It has ideas for coding changes, spots unmentioned small bugs, suggests invariants, principles to adopt, lint rules, tooling tweaks, skill-file updates, follow up work, all kinds of stuff.

View on HN

Memories have mostly been detrimental to me, I have disabled it completely after many moments of frustration.

In general Claude would decide to save irrelevant memories, or apply a memory that was lifted during work on a project to a completely different context where it wasn't relevant, or saved a memory from how to use a tool while I was experimenting with the toll and silently apply it breaking my workflow.

I can't remember a single instance where it was helpful so I just disabled it to not have to deal with yet another cognitive overload.

> I found Opus's 4.8 memories largely lacking value.

Similar thoughts, I disabled memory for the web UI.

> I've been holding "retros" with the agent immediately after a session, and _those_ responses have been typically fantastic.

I used a skill instead to update a ai docs directory as needed.

In any case, I'm moving coding tasks to Codex, Claude will be demoted to frontend design only. I will not entertain the idea of API billing/extra usage or a 2nd Max account once the extra 50% Claude Code usage limits ends on 13 July [1].

[1]: https://news.ycombinator.com/item?id=48126429

For me, the memory capability is a holding zone for deviations from behavior I'm looking for. I've found that a "retro" after to process "memories" and create CLAUDE.md instructions from them instead works more reliably longer term. I don't do it every session. I've got a separate tool which tracks memories on a per-project basis and recommends when to hold that retro based on accumulated memories.

Self-directed memories have been awful.

That said, I have had decent success in self-feedback-loop skills. I encapsulate a long process into a skill and the last step is to make changes as needed to the skill itself for things it ran into. I think the isolation of "this specific thing" is what makes it higher signal. Project or god-forbid global memory is just a recipe for hallucinations in my experience.

Seem to work fine for me but I usually add 1 sentence things to keep the agent from long winded rediscovery.

Remember credentials to query database are in secret store x

Remember company documentation is in platform y

Remember bug fixes always need tests added

Reply on news.ycombinator.com