Context Management
The art of deciding what to keep, drop, and summarize so an agent’s limited context stays useful.
Think of it like
Packing a carry-on — you can’t bring everything, so you choose what earns its space and leave the rest.
Example
As a long agent run fills up, older tool outputs get summarized into a few lines to make room for the current step.
How it actually works
Context is finite and every token costs money and attention, so managing it is one of the central crafts of building agents. Tactics include summarizing old turns, evicting stale tool results, and retrieving only what’s relevant instead of stuffing everything in. Get it wrong and the agent either forgets crucial facts or drowns in noise — and models notably struggle to use the middle of a very full context anyway.
For product teams
Context management is where a lot of agent quality and cost is quietly decided — it deserves real attention.
For engineers
Actively curate the context window each step: summarize, evict, and retrieve to fit the budget without losing signal.
Related
- Context window — The scarce space being managed.
- Memory Retrieval — Deciding what to pull back in.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome