{"slug": "your-ai-remembers-the-old-answer-how-do-you-handle-facts-that-changed", "title": "Your AI remembers the old answer. How do you handle facts that changed?", "summary": "A developer at Wontopos described a memory API that handles outdated facts by letting callers explicitly replace a stale memory via a `supersede` call, which marks the old value as replaced rather than leaving it to be retrieved alongside the new one, and keeps revisions so changes are visible. The engineer also added a `STALE` axis to Glasshouse, a draft open memory benchmark, to measure how well memory systems handle superseded facts, noting that deciding when to call `supersede` remains the developer's responsibility.", "body_md": "I work at Wontopos, so I'm biased. I build a memory API. But this post isn't about it. It's about a problem I keep reading about, and I want to hear how you deal with it.\n\nAn AI that remembers things will sooner or later remember something that is no longer true.\n\nHere is a made-up example, just to show the shape of it:\n\n| When | What the user says | What gets stored | \n|---|---|---|\n| March | \"I live in Seoul.\" | lives in Seoul | \n| July | \"I moved to Busan last month.\" | lives in Busan | \n| October | \"Find me a good dentist nearby.\" | ??? | \n\nBoth memories are true statements. One of them stopped being true. If the memory layer only looks for text that is similar to the question, both come back, and the model has to guess which one wins. Sometimes it guesses wrong, and it sounds confident when it does.\n\nThat's the part that bothers me. A missing memory is easy to notice. An old memory that sounds right is not.\n\nFrom what I've read, people seem to do one of these:\n\nIn our API, an outdated memory is replaced on purpose with a `supersede` call, so the old value is marked as replaced and not just buried. We also keep revisions, so you can see what changed. I don't want to oversell this. Deciding *when* to call it is still your job, and I'm still thinking about how to do that part better.\n\nWe also put a `STALE` axis in Glasshouse, an open memory benchmark I'm working on (still a draft), because I couldn't find an easy way to measure this. Details are in the repo if you want to look.\n\nI'd like to hear from people who build or run their own memory:\n\nYou don't need to sign up for anything. The details are at [wontopos](//wontopos.com), and the site has a short survey on the main page. A comment here is just as good, and the survey is a bonus.", "url": "https://wpnews.pro/news/your-ai-remembers-the-old-answer-how-do-you-handle-facts-that-changed", "canonical_source": "https://dev.to/woochan/your-ai-remembers-the-old-answer-how-do-you-handle-facts-that-changed-1fi7", "published_at": "2026-10-26 08:08:23+00:00", "updated_at": "2026-10-26 08:14:37.457126+00:00", "lang": "en", "topics": ["ai-agents", "ai-tools", "artificial-intelligence"], "entities": ["Wontopos", "Glasshouse"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/your-ai-remembers-the-old-answer-how-do-you-handle-facts-that-changed", "markdown": "https://wpnews.pro/news/your-ai-remembers-the-old-answer-how-do-you-handle-facts-that-changed.md", "text": "https://wpnews.pro/news/your-ai-remembers-the-old-answer-how-do-you-handle-facts-that-changed.txt", "jsonld": "https://wpnews.pro/news/your-ai-remembers-the-old-answer-how-do-you-handle-facts-that-changed.jsonld"}}