Context Engineering
Bağlam Mühendisliği
D3
In this entry, context engineering is the deliberate selection and arrangement at run time of the information and tools supplied in the format needed for an LLM to perform a task.
Review status: 2026-12-04
Technical explanation
For an agent, this can include instructions, message history, available tools, output format, retrieved material, and how state is saved or summarized between steps.
Conceptual boundaries
Context engineering changes what a system presents to a model at run time. It is not synonymous with prompt writing, model training, fine-tuning, memory storage, or a guarantee of factual or safe output.
Provider-neutral example
Illustrative: before asking an agent to draft a policy summary, a system supplies a current policy version, a citation requirement, a response schema, and only the tools needed to locate approved records.
Limitations
Better-selected context can still be incomplete, outdated, conflicting, maliciously supplied, or poorly interpreted. Evaluate outcomes in the intended use context.
Related concepts
Atomic claims and evidence
1.1LangChain Docs: Context engineering in agents
- Source
- LangChain Docs: Context engineering in agents
- Source role
- Authoritative source
- Exact locator
- Overview: ‘Why do agents fail?’
- Supported claim
- LangChain describes context engineering as providing the right information and tools in the right format so an LLM can accomplish a task.
- Last verification
- Review due
- Scope limitation
- This is LangChain’s practitioner definition; it does not establish a single industry-wide method or prove reliability for a particular agent.
2.1LangChain Docs: Context engineering in agents
- Source
- LangChain Docs: Context engineering in agents
- Source role
- Authoritative source
- Exact locator
- ‘What you can control’; ‘Transient context’; ‘Persistent context’
- Supported claim
- LangChain labels model context—what goes into a model call, such as instructions, message history, tools, and response format—as transient; separately, it labels tool context and life-cycle context as persistent information accessed or changed through tools and between steps.
- Last verification
- Review due
- Scope limitation
- These are LangChain’s separate framework categories. The claim does not say that all persistent information is supplied to every model call or that every application stores the same data.