Agentic Context Management: Solving Agent Memory and Cost by Treating Them as Lifecycle and Architecture Problems
Gaurav Dadhich · arXiv · 2026
Abstract
A position paper arguing that production agents fail less from weak reasoning than from mismanaged context — swelling histories, bloated tool definitions, unbounded tool output. It rejects framing this as storage and retrieval, casting it instead as a lifecycle: deciding what to keep, structuring it, choosing a store per data type, forgetting while preserving provenance, and compacting to a budget. Naive accumulation, it argues, grows token cost quadratically with conversation length.
Why it matters
Read this before you build another retrieval layer for your agent. The quadratic cost argument alone reframes context management as an architecture decision rather than a tuning detail.
https://arxiv.org/abs/2607.21503