What is Context Rot?
The gradual degradation of an AI system's effectiveness as the context it relies on becomes outdated, irrelevant, or contradictory. In long-running agent sessions, accumulated context can drift from the original intent, leading to increasingly poor decisions. In RAG systems, stale or poorly maintained knowledge bases produce confident but outdated answers.
My perspective
In practice
Context rot quietly degrades AI-assisted workflows. Mitigation strategies include regular context window resets, versioned knowledge bases with expiry metadata, and explicit "context health" checks built into agent loops. Teams that treat context curation as maintenance (not a one-time setup) get noticeably better AI output over time.