# I don't reread my whole career every morning. Why does my coding agent? — retained record

> Retained reddit record from reddit:ClaudeAI.

Canonical: https://highsignal.app/data/records/abb126b9bc6af6c9

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:ClaudeAI)  published 2026-11-09 17:19 ingested 2026-11-10 02:37

[original source ↗](https://www.reddit.com/r/ClaudeAI/comments/1xrxmcx/i_dont_reread_my_whole_career_every_morning_why/)

## retained text



```
**Disclaimer zero: this is NOT about a product. It's about a broken process. Skip my comment if you like, I'd much rather hear what you think about the questions at the end.**

Disclaimer: written by me, corrected by our friend Claude.

I'm an old(ish) coder. My memory is still good, but I forget small details and decisions, especially when I'm jumping between a lot of projects or one project runs for a long time. Also I hate the fact that every new session starts like a baby I have to teach to eat and write all over again. Every new project? A new nightmare to teach before you can start.

So I did what 30 years of hands-on CTO experience told me to do and RTFM'd Claude first. I started with CLAUDE.md. It grew out of control very quickly, full of stale and contradicting information and I kept getting the dreaded "your CLAUDE.md is too long" warning. Then I split it into a bunch of files all linked from CLAUDE.md. Plus every coding agent wants its own file: CLAUDE.md, AGENTS.md, GEMINI.md, .cursor/rules (bloody standards).

Next I went looking for memory add-ons for Claude and each one was worse than the last. They all behaved like one giant [CLAUDE.md](http://CLAUDE.md) or a pile of md files, with exactly the same problem: they grew without limit and poisoned the context with facts that were irrelevant or contradicted each other. The smarter ones are RAG, which I've always hated: chop everything up, embed the chunks, return whatever lands closest and hope it's still true.

Somewhere in all that I realised the problem isn't any single tool. It's the model of memory they all share: memory as storage. And every 3rd party memory tool I tried? Same thing with nicer packaging. They record EVERYTHING, not what matters, then dump the whole pile into the context on every turn or every search. Bye bye context window, the one my agent actually needs for the code.

My brain doesn't work like that. Nobody's does. I don't reread my whole career every morning. "Staging is down" fades from my head in a day while "we chose Postgres because of X" stays for years. If I change my mind about something, the old idea is gone. It doesn't hang around next to the new one shouting just as loud. And I remember who told me what and more or less when, so I know if I should trust it. I also don't search my memory before I act. The situation itself brings the right thing back: I see the deploy script and remember the gotcha that burned me last time.

A [CLAUDE.md](http://CLAUDE.md) does none of that. It never forgets, never knows which line replaced which, can't tell a decision from a guess and only helps if the model reads the right part at the right moment. Search over old sessions is worse in a way, because it hands back everything that was ever said, dead decisions included. Then the model has to work out which one is still true.

So here's what I keep coming back to. Should an agent's memory forget on purpose? Should a new decision delete the old one, or just outrank it? Should the agent ever have to ask for a memory, or should the right one simply be there when the moment calls for it? And should ALL my coding agents on all my devices know exactly the same facts, decisions and preferences? When I go to work I wear a suit and tie, at home my pyjamas, but I give the same answers and know the same things. Right now Claude on my laptop and Cursor on my desktop are two strangers who've never met.

I've built my own answer to this over the last months and I'll put it in the comments for anyone curious. But first I want to hear what you think.
```



document fetched 2026-11-10 00:22 · canonical https://reddit.com/r/ClaudeAI/comments/1xrxmcx/i_dont_reread_my_whole_career_every_morning_why
