Dictionary
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.
Agentic coding is what happens when you stop asking an AI for code snippets and start giving it goals: "add rate limiting to the API," "migrate this module to the new router." The agent plans the work, makes the edits, runs shell commands, checks the results, and loops until done — or until it gets stuck.
How it differs from vibe coding
Vibe coding is conversational and reactive; agentic coding is delegated and planned. The agent maintains working memory of the task across steps, which lets it handle changes too large to fit in a single prompt. Tools like Cursor's agent mode, Copilot's coding agent, and Claude Code all work this way.
The supervision problem
An agent's output has a larger blast radius than a snippet's. A wrong guess in one file is a bug; a wrong assumption propagated across twelve files is an incident. Agentic coding multiplies leverage in both directions — which is why the review discipline matters more here than anywhere else in AI-assisted development.
Doing it well
Small, verifiable steps. Run the tests after every change. Read the diff before you accept it — all of it. And never let an agent touch infrastructure, auth, or billing code without understanding exactly what it did. The agent is fast; you're responsible.
Related terms
- 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.
- 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.