What wins when an agent's memories conflict? Devs compare write-time vs read-time resolution
MemStrata · reddit · 2026-10-24
A developer running agents with persistent cross-session memory opened a discussion on an under-discussed problem: when two stored facts disagree — a user preferring morning meetings then afternoons, a runbook saying a $5k approval limit while a newer note says $10k — retrieval happily returns both, and something has to decide.
The post surveys the common resolution strategies and asks what actually works in production:
- Recency / last-write-wins
- Source authority, where some entries outrank others
- Keep both and let the model arbitrate at answer time, vs. resolving at write time
The author is especially curious about the losing entry: deleting it destroys the audit trail of what changed and when, but keeping it requires marking it superseded so it stops being retrieved as current fact.
His core argument: retrieval gets all the attention, but the moment two memories disagree, retrieval has already done its job — conflict handling is a separate problem everyone eventually hits and nobody talks about. The thread collects real-world practices from people running memory-backed agents.
More from coding & agent
- uv cuts peak memory another 10-20% for large dependency solves — charliermarsh · 2026-10-24
- 8 rules for running AI agents, each learned from a real failure — petrusenko_max · 2026-10-24
- Cut agent token usage by up to 90%: what RTK, Caveman and Ponytail teach about harness compression — gethackteam · 2026-10-24
- One prompt line cuts LLM output tokens 56% and call costs 33% — gethackteam · 2026-10-24
- Coding agent builder: the hard part is knowing when to act, not being smart — 82% on SWE-bench Verified — BarathKanna · 2026-10-24
- Seven steps from AI tool adoption to an agentic org — most companies are stuck at step one — alex_verem · 2026-10-24