Premium membership
Continue this concept with Premium
Premium includes all published courses, graded answers, concepts and PDF guides. See the price and access terms on the pricing page.
View Premium accessone payment · six months of site access
Advanced
Agent Memory: What an AI Assistant Should Remember, and What It Should Not
Agents use several kinds of memory: working context for the current task, semantic facts about the user or domain, episodic records of past interactions, and procedural knowledge of how to do things. This guide explains each, the product question that comes before storage (should this be remembered at all?), and the controls for provenance, expiry, deletion and memory poisoning.
CONCEPT / THE AGENT LIFECYCLE
RELATED CONCEPTS
PRACTISE THIS CONCEPT
Agentic ProductsWe want our assistant to remember users across conversations. How would you design its memory?→Safety, Security & GovernanceThe ML team wants to use customer conversations to improve our models. What do you need to decide before saying yes?→Agentic ProductsYour team wants to build an AI agent to onboard new suppliers. Should it be an agent, or something simpler?→Agentic ProductsWhat's the difference between a chatbot, an AI workflow and an AI agent?→Agentic ProductsEach step of our agent works 95% of the time in testing, yet users say it fails constantly. Explain, and fix it.→Agentic ProductsOur agent sometimes loops for minutes and spends $4 on a task that should cost 20 cents. What do you do?→