Sahaj.blogs
18 Jul 2025 · Paper summary

Paper summary: A Survey of Context Engineering for Large Language Models

A survey of over 1,400 papers that defines "context engineering": retrieving, processing and managing everything put into an LLM's context, and how RAG, memory, tools and multi-agent systems build on it.

Problem

LLM performance depends heavily on what’s in the context at inference time, and “prompt engineering” is too narrow a name for everything that now goes into building that context.

Method

The survey defines context engineering as the systematic optimisation of the information given to LLMs, with a taxonomy:

  • Foundational components: context retrieval and generation, context processing, context management.
  • System implementations: retrieval-augmented generation (RAG), memory systems, tool-integrated reasoning and multi-agent systems.

It’s based on an analysis of over 1,400 papers.

Key results

Beyond the roadmap, the survey highlights a gap: models with advanced context engineering are very good at understanding complex contexts, but much weaker at generating equally sophisticated, long-form outputs.

Limitations

As a survey it maps existing work. A companion reading list is maintained on GitHub, and the field moves quickly.

#llm#rag#agents#survey

Sources

  1. A Survey of Context Engineering for Large Language Models (arXiv:2507.13334) — arXiv
  2. Hugging Face Daily Papers, 2025-07-18 — Hugging Face
  3. A Survey of Context Engineering for Large Language Models (arXiv HTML full text) — arXiv
  4. Code repository (linked in the paper) — GitHub

This is a summary written from the linked public sources. Figures and numbers are as reported there.