# How do you stop agent memory becoming an outdated source of truth across projects? — retained record

> Retained reddit record from reddit:AI_Agents.

Canonical: https://highsignal.app/data/records/2b5e97c53df7189b

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


source: [reddit](https://highsignal.app/data/reddit) (reddit:AI_Agents)  published 2026-11-14 04:56 ingested 2026-11-15 02:34

[original source ↗](https://www.reddit.com/r/AI_Agents/comments/1xvtpp5/how_do_you_stop_agent_memory_becoming_an_outdated/)

## retained text



```
We use AI across several small businesses with separate web apps and internal systems. We already save context and decisions, so I'm not starting from an agent that forgets everything. What I want to understand is how people keep that memory trustworthy months later.

For example, a saved note might describe an old setup while the current code or account says something different. Or a useful general lesson from one project might get mixed up with facts that only belong to that business.

For people maintaining this in real use, what is your rule for deciding what gets remembered, what expires, and what must be checked again before the agent acts? Do you let the agent update its own permanent memory, or review those changes?

I'm particularly interested in how you handle a contradiction between a remembered decision and a newer source. A concrete failure and the fix would help more than a list of memory tools. What have you found worth keeping, and what did you eventually stop storing?

---

**Edit to make our setup more concrete:** We aren't only building an agent app. We work through coding assistants inside each business's application repo for engineering, content/SEO, social creative, analytics research and admin tasks. We already have separate skills, operating docs and saved decisions, with Obsidian used for curated longer-term notes. The question is how to maintain the boundaries between those sources as work and tools change.

One actual contradiction I found while reviewing the setup: a dated audit says there are no hooks, but the current repository contains hooks. That old audit still has useful historical context; it just isn't a current inventory. A paused or superseded workflow can create a similar problem if the agent treats the old plan as permission to run it now.

So what I need is a practical way to distinguish current instructions, current task state, historical decisions and reusable lessons across many kinds of work. Where would you put the Obsidian notes in that hierarchy, and what causes you to review or retire a note? I'm not proposing that every business fact should go into a shared memory pool.
```



document fetched 2026-11-15 00:18 · canonical https://reddit.com/r/AI_Agents/comments/1xvtpp5/how_do_you_stop_agent_memory_becoming_an_outdated
