---
schema_version: "newruntime-agent-readable-v0.1"
type: "trend_pattern"
slug: "company-memory-needs-write-loops"
title: "Company memory needs write and correction loops"
description: "Retrieval is only one stage of memory; durable organizational knowledge also needs scoped writes, corrections, provenance, versioning, and invalidation."
thesis: "Company memory becomes useful when real work continuously updates a governed knowledge layer, not when an agent merely searches a larger document pile."
status: "published"
confidence: "high"
first_seen: "2026-06-24"
last_verified: "2026-07-16"
supporting_signals: ["tg-2703","tg-2695","tg-2625","tg-2525","tg-2631"]
related_posts: ["shared-document-human-agent","unabyss-shared-context-layer"]
counter_evidence: ["For stable reference material, ordinary retrieval over maintained documents can be sufficient and cheaper than a dedicated memory lifecycle.","Automatic memory writes can amplify mistakes, stale assumptions, privacy violations, and unauthorized knowledge sharing."]
revision_trigger: "Revise the thesis if retrieval-only systems match governed write-loop systems on repeated correction rate, freshness, provenance, access control, and rollback quality."
topics: ["company-memory","agent-memory","knowledge-systems"]
source_urls: ["https://promptql.io/","https://aclanthology.org/2024.acl-long.747","https://github.com/OpenDataBox/MemoryData","https://careersatdoordash.com/blog/doordash-unified-consumer-memory-for-personalization-at-scale","https://github.com/langchain-ai/openwiki"]
routes: {"html":"https://newruntime.com/patterns/company-memory-needs-write-loops/","markdown":"https://newruntime.com/patterns/company-memory-needs-write-loops.md","json":"https://newruntime.com/patterns/company-memory-needs-write-loops.json"}
source_format: "markdown"
---

# Company memory needs write and correction loops

## What retrieval leaves unresolved

A relevant document can still be stale, unauthorized, contradicted, or scoped
to another team. A long context window can expose more text without deciding
which statement should change the next action. Those are memory lifecycle
questions, not search-quality questions.

## Supporting evidence

- PromptQL captures a correction during work, proposes a reusable rule, and
  keeps citations, scope, revision history, and editorial control around the
  update.
- DoorDash stores semantic memory as independently versioned components and a
  manifest, then refreshes different memory layers at different cadences.
- MemoryData evaluates storage, retrieval, updating, and forgetting across a
  sequence rather than scoring only a final response.
- OpenWiki turns repository understanding into a maintained artifact that can
  be inspected and refreshed outside an individual session.

## Operational consequence

Every memory entry needs a write path, evidence, owner or authority, scope,
timestamp, supersession behavior, and rollback. Human correction is a valuable
write signal, but it should produce a reviewable proposal rather than an
unbounded automatic truth update.