I stopped treating CLAUDE.md like memory now that Claude can actually remember things

Published Sep 18, 2026, 7:30 AM EDT Mahnoor Faisal is a tech journalist covering AI and productivity tools with bylines at XDA, SlashGear, MakeUseOf, Laptop Mag, and Android Police. She's been writing professionally since she was sixteen, and has since penned hundreds of articles. This includes in-depth coverage of AI tools like NotebookLM to breaking news across the AI space. Her passion for technology started when she received her first iPod Touch (4th generation) on her 8th birthday, and she's been deep in the tech world ever since. Currently pursuing a degree in computer science, Mahnoor brings both a journalist's eye and a technical foundation to her coverage of how AI is reshaping the way we work and learn. You might've noticed that the AI giants have, at least somewhat, slowed down their habit of launching new models left and right over the last few months. That's relative, of course. New models are still arriving constantly, but compared to the breakneck pace we got used to, it feels like a lot more attention is being directed toward improving the actual experience around them. Beyond squeezing more intelligence out of each model, AI labs have been putting much more effort into making their tools feel less disposable from one conversation to the next. OpenAI has continued expanding ChatGPT's memory, Google has been leaning heavily into personalization and memory with Gemini, and Anthropic has been doing the same across Claude. This means that instead of constantly trying to keep up with whatever new model just dropped, I've found myself paying a lot more attention to features that have quietly been sitting there for a while. Claude Code's auto-memory is one of them. It isn't particularly new at this point, but I only recently realized just how much of my CLAUDE.md setup it could take off my hands (and how much I had been overcomplicating my own workflow without really needing to). Claude Code has two ways to remember things Not all memory is created equal Similar to other AI tools, Claude's web version has a memory feature that lets it carry useful context about you across conversations instead of starting completely from scratch every time. Claude Code has always worked a tad bit differently from the very beginning, though. Initially, the only native way to give Claude Code persistent context across sessions was through files you maintained yourself. The most obvious place to put everything you wanted the tool to keep in mind was CLAUDE.md, since this was a file Claude would automatically read at the beginning of every session and use as context throughout your work. In February, that changed. Claude Code v2.1.32 introduced auto-memory, allowing Claude to automatically record and recall useful information as you work with it instead of relying entirely on whatever you remembered to write down yourself. Claude Code's support documentation describes auto-memory and CLAUDE.md as two "complementary memory systems," with both loaded at the beginning of every conversation. Auto Memory handles the things I used to manually preserve One less thing for me to babysit While auto-memory and CLAUDE.md are effectively two distinct ways to give Claude persistent context, the biggest differences between them are who does the remembering and what kind of information each system is meant to preserve. With the CLAUDE.md file, I'm the one deciding what Claude should know and manually writing it down. In my case, given that I have a very specific way of working and tend to have different preferences depending on the project, that meant my CLAUDE.md files kept growing over time as I added more instructions, corrections, and little bits of context I didn't want to have to repeat in every new session. Auto Memory is maintained by Claude and Claude only. As you continue to use it, Claude picks up on useful learnings and patterns from the way you work, including preferences you repeatedly express, corrections you make, and project context it can't simply infer from the code itself. It then writes those down automatically so they can carry over into future sessions, without you having to stop and decide whether each little detail deserves a spot in CLAUDE.md. Before Auto Memory launched, I stored everything it now picks up automatically in CLAUDE.md myself. If Claude kept getting something wrong, I added a note. If I realized I preferred a certain workflow, that went in too. The same applied to project-specific quirks, recurring corrections, and pretty much anything else I thought might save me from having to explain myself again in a future session. While the CLAUDE.md file certainly still has its place, looking back, I was making it do a lot of work that Claude can now handle on its own. With the feature now in place, Claude can learn from the corrections I make and the preferences I express as I work, then carry those lessons into future sessions without me having to manually preserve every single one! I now keep CLAUDE.md for the rules that really matter Back to its actual job As I mentioned above, the CLAUDE.md file absolutely still has its place. In fact, I'd argue that the auto-memory feature has only made its purpose clearer. Instead of using it as a dumping ground for every little preference or correction I want Claude to remember, I now reserve it for the instructions and project details I explicitly want Claude to have in front of it every single time. For example, if I'm working on coding a management system, instead of adding my own coding preferences, the way I structure functions or corrections I've made along the way, I can reserve the CLAUDE.md file for information that's specific to the project itself. That could include its architecture, important commands, naming conventions, database constraints, or any rules I need Claude to follow consistently throughout the project! This single file is still an extremely important part of my coding workflow, but paired with auto-memory, it no longer has to carry the entire burden of making Claude feel consistent from one session to the next!