{
  "id": 9912666,
  "title": "Making an AI remember is harder, and worth more, than making it smarter",
  "url": "https://urgent.news/2026/09/26/making-an-ai-remember-is-harder-and-worth-more-than-making-it-smarter",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-26T04:42:00.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/derekwang85/making-an-ai-remember-is-harder-and-worth-more-than-making-it-smarter-1m00"
  },
  "original_language": "en",
  "account": "Derek Wang, author of Borges wrote a terrifying gift into Funes the Memorious, argues that engineering AI systems to remember is far more challenging than making them smarter. In his essay, Wang delves into the paradox that forgetting is essential for thought, as a mind that cannot forget cannot think.\n\nAI harness engineering has focused on steering AI-generated content, restraining where it may lead, but the one thing that truly determines an AI system's long-term survival is its ability to remember what it did yesterday. Each conversation is a fresh amnesia for AI, similar to how a human retelling of an incident leads to distorted details over time.\n\nThe scenario of an AI forgetting information is detrimental, as it requires repeating explanations and distorting details with every retelling. Wang notes that AI's every session resets the context window, making it feel like the new hire on the first day. This leads to the essay's central thesis: eight essays are built around the concept of amnesia in AI systems, highlighting the need for long-term memory.\n\nWang introduces gbrain, a knowledge system that held project knowledge, historical decisions, and agent experience. Although gbrain initially seemed complete, the migration to derekinside revealed that knowledge barely changed by a single word. The true difference lay in retrieval paths: gbrain forced human memory retrieval, while derekinside automatically retrieved information, significantly reducing latency from 5-10 seconds to under one second.\n\nDerekinside made the knowledge base more accessible and efficient, transforming it from a dead exhibit of numbers to a system that automatically looks up information. The key to derekinside's success lies in retrieving information first, then thinking, enforced by the AGENTS.md file that mandates every agent to check derekinside before proceeding.\n\nThe essay concludes by emphasizing that memory must be maintained, not just installed. A real-world example illustrates that even with a well-structured knowledge base, a system's performance can suffer if retrieval is not properly maintained. When the embedding model was swapped, the index became stale, causing semantic search to fail. The solution was to rebuild the index and restore retrieval, reinforcing the notion that memory is not a one-time installation but an ongoing process.",
  "summary": "AI Harness Engineering · Essay Nine · derek wang (derekwang85) Borges wrote a terrifying gift into Funes the Memorious . After Funes falls from a horse, he acquires a memory that never forgets — the outline of every leaf on every morning, the wrinkles on every face he saw in a day. He becomes a living archive so precise it borders on absurd. But he does not become smarter. He drowns. Because he…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}