What is context engineering?

Context engineering is the practice of giving an AI model the right information, instructions, and tools for a task, rather than relying on the prompt alone. For coding agents it means repo-level files like CLAUDE.md that cut wrong assumptions.

Context engineering vs prompt engineering

Prompt engineering focuses on wording a single request well. Context engineering is broader: it assembles everything the model needs to act, including the relevant files, the rules of the codebase, and the tools it can call. A sharp prompt still fails when the model lacks the surrounding facts.

• Prompt engineering shapes the instruction.
• Context engineering shapes the information, instructions, and tools around it.
• The two work together, and context sets the ground the prompt stands on.

What goes into context

Context is the full set of inputs an AI model reads before it responds.

• Instructions: the task, constraints, and conventions to follow.
• Information: source files, documentation, and prior decisions.
• Tools: the commands, searches, and APIs the model can call.
• State: results from earlier steps in a multi-step task.

Why it matters for coding agents

A coding agent works across many files and makes changes on its own. When it cannot see the conventions of the surrounding code, it fills the gaps with assumptions, and a confident assumption becomes a defect in the diff. Strong context removes those assumptions before they reach the change, which matters more as agentic coding hands more of the work to the model.

Practical patterns

Teams supply context to AI models in a few repeatable ways.

• Instruction files: a repo-level file such as CLAUDE.md states the stack, conventions, and rules every task should follow.
• Retrieval: the system pulls the most relevant files into context instead of loading the whole repository.
• Tool definitions: clear descriptions of what each tool does, so the model calls the right one.
• Scoping: limiting context to what the task needs keeps the model focused.

A shared standard such as the Model Context Protocol gives tools a common way to supply this context, and the same discipline improves any AI code generation workflow.