AI agents and technology
What is AI memory? Definition and examples
Also: Agent memory, session memory, AI memory layer
In short
AI memory is the part of an AI system that keeps knowledge beyond a single conversation. Short-term memory holds the current task; long-term memory stores facts and decisions for later. Unlike RAG, which looks things up in documents, memory keeps what the system has learned.
Why it matters: Memory saves you from explaining the same things to every AI agent every day.
- Short-term memoryThe current task in the context window
- Long-term memoryDecisions and facts across conversations
- Documents via RAGLooked up when needed, not remembered
Why does AI memory matter for your company?
Language models forget everything after each conversation. Without memory every task starts from zero: you explain roles, goals and decisions again and again. Memory turns single chats into ongoing collaboration.
Short-term memory lives in the context window and disappears with the task. Long-term memory sits in a store the system reads from later when needed. RAG complements both: it finds relevant passages in your documents for each question, but remembers nothing new.
AI memory in detail
Agent memory
Agent memory is what a single AI agent keeps from one run to the next: results, instructions and earlier decisions. It no longer has to understand a task from scratch every time.
The limit: each agent learns on its own. Without a shared, approved state, three agents keep three versions of the same instruction and sooner or later contradict each other.
Session memory
Session memory is the short-term memory of a single session with an AI. Everything clarified in the chat only holds until that session ends.
In the next session the AI starts from zero again, and the team explains goals and rules once more. Lasting memory needs a store outside the session.
AI memory layer
An AI memory layer is a separate store alongside the language model. It records what was settled in conversations with AI and hands it on to later sessions and other tools.
For a company the rule behind it matters more than the technology: who decides what applies, and who approves changes? Without that rule the store also keeps what is outdated or contradictory.
Example: The assistant that remembers the wrong thing
A head of product has used an AI assistant with memory for months. The assistant has remembered that the US expansion is a priority for 2027. Then leadership decides “Strategy 2027: focus on the DACH mid-market, no US expansion”, yet the assistant keeps proposing US features for the roadmap.
The problem is not forgetting but remembering the wrong thing. The team records the new decision in one shared place all assistants read and deletes the outdated entry. From then on every assistant remembers the same valid state.
3 common mistakes to avoid
Confusing memory with RAG
RAG finds what is in documents. Memory keeps what was decided or learned along the way. You usually need both.
A private memory for everyone
When every assistant remembers something different, private truths appear. Company knowledge belongs in a shared, maintained memory.
Never forgetting anything
Memory without upkeep also keeps what is outdated. Entries need a source, a date and a person allowed to change them.
Step by step
- 1
Decide what should last
Decisions, goals and fixed terms belong in long-term memory; interim steps do not.
- 2
Choose one shared place
All assistants read from the same memory instead of private notes.
- 3
Store source and date
Every person and agent can then see where an entry comes from and how old it is.
- 4
Replace what is outdated
When a decision changes, replace the old entry instead of adding a new one next to it.
Bring one important decision.
In 30 minutes with Marc or Simon we show where your work drifts from the decision today.
Meet the Company BrainRead on in the PM Lab
Articles that go deeper.
Related terms
- RAG →RAG (retrieval-augmented generation) is a method where an AI model looks up relevant passages in your documents before it answers.
- Context window →The context window is the amount of text an AI model can read at once when it answers.
- Organisational memory →Organisational memory is what a company knows and keeps about itself: decisions, experience and reasons, independent of individual people.
- Company Brain →A Company Brain is a company's shared memory: it connects knowledge, decisions and meeting notes from the tools you already use in one place.