AI Memory and Curated Context
What an agent should know about your organisation before it acts: memory, context, and the gap retrieval cannot close.
11 answers
- What is AI agent memory?'Does this agent have memory?' is three different questions. Session memory, retrieved context, and organisational memory are different layers, different problems, and solved by different tools. Most current tooling covers the first two.
- What is organisational memory for AI?Personal memory tracks users. RAG retrieves documents. Neither captures the reasoning behind decisions that was never written down. What organisational memory for AI means, and what breaks without it.
- What's the difference between AI memory and RAG?RAG fetches from a corpus. Memory persists across interactions. Neither captures what an organisation has deliberately decided before its agents act. The difference, and the gap both leave open.
- Do I need a vector database for my AI agents?Probably, for retrieval. But a vector database stores and retrieves whatever you put in; it doesn't capture what your organisation has decided or what constraints its agents should operate under. Two different problems.
- What is the difference between AI context and AI memory?Context is what an agent is given for a task. Memory is what persists across sessions. They fail differently, and improving one doesn't fix the other. The upstream question both leave open.
- What is demand-driven context for AI agents?Retrieval surfaces what exists and is relevant. Demand-driven context starts from what the agent actually needs, and makes the gaps visible before execution begins, not after.
- What is temporal continuity in AI?A knowledge base records what the organisation knows. Temporal continuity records how it came to know it: what changed, when, and why. The difference shows up at every boundary case an agent encounters.
- What is institutional knowledge in the context of AI agents?Institutional knowledge is what experienced people know that isn't in any document. When agents are deployed to do work those people previously did, the gap doesn't disappear; it becomes invisible.
- What is context engineering and is there a better alternative?Context engineering is the practice of designing what goes into an agent's context window. Most of the effort is compensating for an absent upstream authorship layer. What changes when that layer exists.
- Where does my organisation's AI context actually live?Most organisations have AI context distributed across at least five tools, partially in provider runtimes, not designed for inspection, and non-transferable. The question nobody asked clearly enough when the tools were adopted.
- What did the agent know before it acted?The logs show what the agent did. They rarely show what it was working from. The difference between retrieving that record and reconstructing it is determined before the agent ever acts.