Why can language models understand context better than generate it?
Models absorb and process rich input context far more effectively than they produce similarly sophisticated outputs. Understanding this asymmetry could reshape how we design systems to compensate for generative limitations.
LLM performance is fundamentally governed by the context supplied at inference, and this survey of over 1,400 papers argues the methods for designing that context have matured into a formal discipline — context engineering — that transcends prompt design. Its taxonomy decomposes into foundational components (context retrieval and generation; context processing for long sequences, self-refinement, and structured integration; context management covering memory hierarchies and compression) and their architectural implementations (RAG, memory systems, tool-integrated reasoning, multi-agent orchestration).
The keeper is not the taxonomy but the central open challenge it surfaces: a comprehension-generation asymmetry. Models are remarkably good at understanding complex contexts yet limited in generating outputs of comparable sophistication. The lever that improves a model's behavior (richer, better-organized input context) operates on a different and stronger axis than the model's own generative ceiling. This reframes "make the model better" toward "engineer the information payload" — and explains why so much practical capability now comes from context rather than weights.
This gives the vault's context cluster a unifying frame. Since Can context playbooks prevent knowledge loss during iteration?, the playbook view is one implementation of the management component; and the comprehension-generation gap is the deeper reason Where does agent reliability actually come from? works — the harness feeds comprehension, which the model does well.
Inquiring lines that use this note as a source 2
This note is a source for these synthesized inquiries. Follow a line forward into its question, or open it to trace back to all of its sources.
Related concepts in this collection 3
This note in its neighbourhood — explore the map, then jump to a related concept in the list below.
Click a node to walk · click center to open · click Open in graph to see this note in the full knowledge graph
-
Can context playbooks prevent knowledge loss during iteration?
When AI systems iteratively refine their instructions and memories, do structured incremental updates better preserve domain knowledge than traditional rewriting? This matters because context degradation undermines long-term agent performance.
one implementation of the management component this survey formalizes
-
How do LLMs balance remembering context versus keeping it separate?
LLMs face a structural tension: retaining too much context causes different threads to blur together, while retaining too little causes the model to lose track of earlier commitments. This explores whether this dilemma is fundamental to how transformers work.
the context-management problem this discipline organizes
-
Can external managers compress context better than frozen agents?
Explores whether offloading context management to a trained external system can adapt compression strategies to individual agent strengths, rather than forcing agents to manage their own context constraints.
a concrete context-management technique under the same discipline
Related papers in this collection 8
Papers most semantically related to this note, ranked by cosine similarity in the embedding space.
- Context Engineering 2.0: The Context of Context Engineering
- A Survey of Context Engineering for Large Language Models
- Explain-Query-Test: Self-Evaluating LLMs Via Explanation and Comprehension Discrepancy
- Can Large Language Models Understand Context?
- Agentic Context Engineering: Evolving Contexts for Self-Improving Language Models
- Generative Models as a Complex Systems Science: How can we make sense of large language model behavior?
- Provable Benefits of In-Tool Learning for Large Language Models
- Between Circuits and Chomsky: Pre-pretraining on Formal Languages Imparts Linguistic Biases
Original note title
context engineering is a formal discipline whose core finding is a comprehension-generation asymmetry — models understand rich context better than they can generate it