Vibeleaderboard
← All Intel
Intel / video

Build Hour: Agent Memory Patterns

Source
youtube.com
Author
OpenAI
Date

Context

A solutions architect at OpenAI says in a Build Hour session that long-running agents are limited by a finite : instructions, history and tool results all compete for the same space. The presenter groups remedies into three patterns: reshape and fit (trim, compact or summarize), isolate and route (give each only the context it needs), and extract and retrieve (save memories for later sessions).

The talk names four ways context goes wrong: a sudden burst of tool output, conflicting instructions, an incorrect fact that is summarized and carried forward, and noise from many overlapping tool definitions. The stated goal is the smallest context with the most useful signal. These are presented as recommended practice, shown in a demo support .

Terms in this piece · Glossary
  • context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
  • subagent — A separate agent spawned by another to do one scoped piece of work in its own context, returning only the result.
  • AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
  • context engineering — Deciding what an agent sees at each step — the discipline that replaced prompt engineering once context windows got large enough to fill badly.
Read the source www.youtube.com
More from OpenAI
Recommended reads
Comments

Checking sign-in…

Loading comments…