Context rot
Context rot is the degradation of a model’s ability to use contextual information effectively as that information becomes excessive, stale, irrelevant, noisy, or conflicting.
Vibe Engineering
Context rot is the degradation of an AI agent's performance as an overloaded or stale conversation context accumulates irrelevant or conflicting information.
- How it is used
- Identifies excessive context as a controllable source of AI agent failures within context engineering.
- How deep it goes
- Explains length and positional effects, offers a utilization heuristic, and describes file-backed workflows as an alternative to compaction.
- Examples
- Uses a UserProfileCard prompt with a buried instruction and the GSD workflow with a CONTEXT.md decision file.
- What you can do after
- Evaluate context overload and distinguish length-related degradation from positional retrieval failures.
Context Engineering
Context rot is degradation in the usefulness of accumulated context as a task evolves and its information loses relevance, structure, or priority.
- How it is used
- Diagnoses loss of useful context in long conversations, multi-step agent runs, and instruction maintenance.
- How deep it goes
- Defines the degradation, describes its symptoms, and names mitigations without specifying implementation procedures.
- Examples
- Uses long conversations, multi-step agent runs, and obsolete commands or paths as examples.
- What you can do after
- Diagnose context degradation and apply pruning, explicit task state, focused summaries, or directive reviews.
Also covered in
- Machines that ThinkIn early access (MEAP)
- Build an AI Agent (From Scratch)
What to know first
- context windowA context window is a bounded span of sequence elements or contextual information that a model or processing operation can consider together.
- language modelA language model is a computational model that represents patterns in linguistic sequences to assign probabilities, predict, generate, or evaluate language.
Related concepts
- stale datacontrasts with · Stale data is information that no longer reflects the latest state of its source or relevant conditions.
- accuracyproduces · Accuracy is the degree to which a result agrees with a specified truth, reference, or criterion, commonly measured as the proportion of evaluated predictions that are correct.
- context failurekind of · Context failure is an insufficiency or mismatch in contextual information that prevents a system from adequately interpreting intent, assessing relevance, or performing a task.
- context pruningaddressed by · Context pruning is the selective removal or summarization of supplied context before a model call, often to reduce its size or retain more relevant information.
- stateaddressed by · State is information describing the condition or configuration of a system or computation at a particular time that may influence its behavior or subsequent transitions.
- state externalizationaddressed by · State externalization is an agent-workflow pattern that persists project information outside ephemeral conversation sessions in durable artifacts such as files.
Questions about context rot
What is context rot?
Context rot is the degradation of a model’s ability to use contextual information effectively as that information becomes excessive, stale, irrelevant, noisy, or conflicting.
Which Manning books explain context rot?
Vibe Engineering and Context Engineering.
What should I know before learning context rot?
Manning books that teach context rot usually introduce context window and language model first.
What is the difference between context rot and stale data?
Context rot is the degradation of a model’s ability to use contextual information effectively as that information becomes excessive, stale, irrelevant, noisy, or conflicting. Stale data is information that no longer reflects the latest state of its source or relevant conditions.
What is context rot a kind of?
Context rot is a kind of context failure. Context failure is an insufficiency or mismatch in contextual information that prevents a system from adequately interpreting intent, assessing relevance, or performing a task.
These guides are generated from the text of Manning books by Manning's knowledge graph and link to the sections they draw on. The full books are available on liveBook and at manning.com.

