Memory: What Agents Should Remember (and Forget)

{
  "byte_id": "agentic_dev_013",
  "title": "Memory: What Agents Should Remember (and Forget)",
  "level": "foundation",
  "domain": [
    "agentic-development",
    "memory",
    "rag"
  ],
  "intent": "Understand different types of agent memory and how to use them without creating confusion, drift, or privacy risks.",
  "core_idea": {
    "one_liner": "If an agent remembers everything, it understands nothing.",
    "why_it_matters": [
      "Uncontrolled memory leads to inconsistent behavior.",
      "Most agent bugs come from stale or misused memory.",
      "Memory must be intentional and scoped."
    ]
  },
  "definition": {
    "agent_memory": "Persisted information that influences future agent behavior beyond the current request."
  },
  "memory_types": [
    {
      "type": "Session memory",
      "purpose": "Maintain context within a single conversation or task.",
      "lifetime": "Minutes to hours",
      "example": "Clarifications given earlier in the same task"
    },
    {
      "type": "Task memory",
      "purpose": "Remember intermediate results during a multi-step workflow.",
      "lifetime": "Until task completion",
      "example": "Collected metrics during RCA"
    },
    {
      "type": "Long-term knowledge (RAG)",
      "purpose": "Store stable facts, rules, and procedures.",
      "lifetime": "Days to months",
      "example": "Decision rules, checklists, anti-patterns"
    },
    {
      "type": "Preference memory",
      "purpose": "Remember user or org preferences.",
      "lifetime": "Explicitly defined",
      "example": "Preferred output format, approval thresholds"
    }
  ],
  "what_should_not_be_memory": [
    "One-off errors",
    "Temporary system states",
    "Speculative conclusions",
    "Sensitive personal data"
  ],
  "memory_write_rules": [
    "Memory writes must be explicit, not implicit.",
    "Every memory item must have a scope and expiry.",
    "Only validated information can become long-term memory."
  ],
  "micro_example": {
    "scenario": "Agent investigates replication delay",
    "bad_memory": "Replication is slow in system X (no context, no date).",
    "good_memory": {
      "fact": "On 2026-01-12, replication delay in system X was caused by queue backlog.",
      "scope": "incident-specific",
      "expiry": "7 days"
    }
  },
  "failure_modes": [
    "Memory pollution (mixing facts and guesses)",
    "Stale memory reused as truth",
    "Hidden memory influencing answers",
    "Unbounded memory growth"
  ],
  "guards": [
    "Memory entries must be reviewable.",
    "Long-term memory must be versioned.",
    "Agents must explain when memory influenced an answer."
  ],
  "teach_it_in_english": {
    "simple_explanation": "Memory is useful only when you know why you remember something.",
    "one_sentence_definition": "Good memory makes agents consistent; bad memory makes them unreliable."
  },
  "practical_checklist": [
    "Why is this being remembered?",
    "For how long is it valid?",
    "Could this memory mislead the agent later?",
    "Can we delete or expire it safely?"
  ],
  "tags": [
    "agent-memory",
    "rag",
    "state-management",
    "knowledge-hygiene"
  ],
  "meta": {
    "schema": "dkharlanau.dataset.byte",
    "schema_version": "1.1",
    "dataset": "agentic-bytes",
    "source_project": "cv-ai",
    "source_path": "agentic-bytes/agentic_dev_013.json",
    "generated_at_utc": "2026-02-03T14:33:32+00:00",
    "creator": {
      "name": "Dzmitryi Kharlanau",
      "role": "SAP Lead",
      "website": "https://dkharlanau.github.io",
      "linkedin": "https://www.linkedin.com/in/dkharlanau"
    },
    "attribution": {
      "attribution_required": true,
      "preferred_citation": "Dzmitryi Kharlanau (SAP Lead). Dataset bytes: https://dkharlanau.github.io"
    },
    "license": {
      "name": "",
      "spdx": "",
      "url": ""
    },
    "links": {
      "website": "https://dkharlanau.github.io",
      "linkedin": "https://www.linkedin.com/in/dkharlanau"
    },
    "contact": {
      "preferred": "linkedin",
      "linkedin": "https://www.linkedin.com/in/dkharlanau"
    },
    "canonical_url": "https://dkharlanau.github.io/datasets/agentic-bytes/agentic_dev_013.json",
    "created_at_utc": "2026-02-03T14:33:32+00:00",
    "updated_at_utc": "2026-02-03T15:29:02+00:00",
    "provenance": {
      "source_type": "chat_export_extraction",
      "note": "Extracted and curated by Dzmitryi Kharlanau; enriched for attribution and crawler indexing."
    },
    "entity_type": "agentic_byte",
    "entity_subtype": "level:foundation",
    "summary": "Understand different types of agent memory and how to use them without creating confusion, drift, or privacy risks."
  }
}