Dictionary
Context engineering
The practice of deliberately assembling what an AI coding assistant knows — relevant files, docs, constraints, and conventions — so its output matches your project's reality instead of generic best guesses.
Context engineering is the unglamorous skill behind good AI-assisted coding: deciding what the model gets to see. An AI assistant with your repo's conventions, the relevant modules, and your constraints will write code that fits. The same assistant with no context writes generic code that looks right and integrates poorly.
What good context looks like
- The files being changed, plus the ones they depend on
- Project conventions (a
AGENTS.md, contributing guide, or style doc) - Constraints stated plainly: "no new dependencies," "must stay under 100ms," "follow the existing error-handling pattern"
- Relevant docs or API references, not the entire documentation site
Why it matters more than prompting
Most "the AI wrote bad code" stories are actually "the AI was given bad context" stories. Prompting tricks fade in importance as models improve; context quality doesn't. The developers getting the best results from AI tools in 2026 aren't better prompters — they're better curators of what the model knows.
The craft angle
Context engineering is just good engineering communication, applied to a new colleague who happens to be a language model. Write down your conventions. Keep docs current. Make the codebase explain itself. Everything that helps a human onboarding helps the AI too — which is a convenient excuse to finally do the documentation you've been postponing.
Related terms
- Agentic coding
A workflow where an AI coding agent plans and executes multi-step software changes — editing files, running commands, and iterating — with a human supervising rather than typing.
- Vibe coding
Writing software by describing what you want to an AI in natural language, accepting its output with minimal review, and iterating on vibes rather than specifications.