Part 1 — Why Agents Need Memory (Foundations)

This section establishes why persistent memory is a first-class problem for agent systems — not a prompt tweak, and not the same thing as a larger context window.

What you will learn

  • Why language models are stateless by default, and what that breaks across turns and sessions
  • How context windows differ from memory, and why longer context alone fails
  • The cost, latency, and trust trade-offs of re-sending history instead of recalling it
  • What personalization, multi-agent coordination, and continual learning actually require architecturally

Chapters in this section

There are 20 chapters in this part. Open any chapter to read it on its own, or work through them in order.

  1. 1 The Statelessness Problem: Why Language Models Forget Everything
  2. 2 The Context Window Is Not Memory
  3. 3 Lost in the Middle: Why Long Context Degrades Recall
  4. 4 Working Memory vs. Long-Term Memory: The Core Distinction
  5. 5 Latency and Cost: The Hidden Price of Re-Sending History
  6. 6 Why “Just Use a Bigger Context Window” Doesn’t Solve Memory
  7. 7 The Difference Between Retrieval-Augmented Generation and Agent Memory
  8. 8 From Chatbot Amnesia to Continuous Identity: What Persistent Memory Enables
  9. 9 Memory as Infrastructure, Not an Afterthought
  10. 10 The Cost of Repeated Reasoning: Why Agents Without Memory Waste Tokens
  11. 11 Single-Turn vs Multi-Turn vs Multi-Session Systems
  12. 12 What “Personalization” Actually Requires Architecturally
  13. 13 Continual Learning Without Retraining: Memory as a Substitute for Fine-Tuning
  14. 14 The Agent Memory Problem in Multi-Agent Systems
  15. 15 Memory and Trust: Why Predictable Recall Matters More Than Total Recall
  16. 16 Noisy, Contradictory, and Time-Varying Data: The Raw Material of Memory
  17. 17 Why Naive Conversation Logging Fails as a Memory Strategy
  18. 18 The Three Questions Every Memory Architecture Must Answer (What, When, How)
  19. 19 Memory as a Competitive Advantage in Agentic Products
  20. 20 A Brief History: From Session Storage to Vector-Native Memory