Everything is Context: Agentic File System Abstraction for Context Engineering
Xiwei Xu, Robert Mao, Quan Bai, Xuewu Gu, Yechao Li, and Liming Zhu; submitted 2025-12-05.
Everything is Context proposes a filesystem abstraction for context engineering: a way to give agent work stable memory, tool grounding, and provenance. This page keeps the paper, the questions it has actually raised, and the formal responses in one place.
The paper treats context as a system concern: resources an agent can find, ground its tools in, and trace over time. It does not present cognition as a filesystem problem or claim to settle every coordination question.
Xiwei Xu, Robert Mao, Quan Bai, Xuewu Gu, Yechao Li, and Liming Zhu; submitted 2025-12-05.
The paper moves context out of the prompt alone and into a system of named resources, tools, and working state.
An agent's work needs resources it can find, retain, and use again, rather than relying only on a one-time prompt.
Tools are more useful when the resources and task they act on have an explicit place in the system.
When work continues across time or participants, the resources and process behind a result should remain traceable.
The Local FS material and Small World article show how the paper's abstraction connects to a local directory and a projected resource view.
These exchanges clarify the paper's scope through questions raised in public.
Karim C, 2025-12-08: how should context drift be handled when multiple agents write to the same directory?
A public question from Turner NextGenAI and Robert's immediate reply: AFS concerns context, not cognition. The original exchange is linked here; no formal response article has been added.
A public question from vbridgesruiz and Robert's immediate reply about presenting context in an interface. The original exchange is linked here; no formal response article has been added.
The Chinese and English response article preserves the original question and immediate reply, then explains the boundary between resource projections and reconciliation.