Context (AI Agent)
The specific information an agent knows and can act on at a given moment.
Context is the body of information an AI agent has immediate access to and considerspertinent to its current task. It is the abstract knowledge set the agent draws from — not the raw tokens fed to the model (that is the context window), not the accumulated history (that is the session), but the active slice of relevant information the agent is working with at a given moment.
Mechanically, context is constructed by what gets loaded into the context window: pasted files, function definitions, prior conversation turns, tool results, and external reads all compete for inclusion. Context engineering is the discipline of curating what enters that window so the agent's attention is directed toward the highest-value information.
The central tradeoff is breadth versus depth. Loading more information increases the chance the agent has what it needs, but dilutes each individual piece of information's influence through the attention budget. Context that is too broad produces unfocused responses; context that is too narrow causes the agent to hallucinate or guess at missing information.
Open questions include how to formally measure context quality and relevance, whether memory systems can dynamically curate context better than manual engineering, and how to handle context that contradicts itself across different loaded sources. Practitioners currently rely on heuristics and trial-and-error rather than principled frameworks.