Agentic Coding

Agentic coding is when an AI tool goes beyond answering questions and starts taking autonomous, multi-step action on your project: reading your codebase, planning changes, editing files, running tests, and iterating on the results without needing you to direct every step. Unlike simple code generation where you prompt and receive, agentic coding gives the AI access to your full project and the ability to make coordinated changes across multiple files and sessions, guided by persistent context like project plans and instruction files.

The Simple Version

Regular AI chat is like texting a friend for advice. You ask a question, they answer. You ask another, they answer again. Each exchange stands alone.

Agentic coding is like hiring that friend as a contractor. They read your entire project, understand the structure, plan the work, make changes across multiple files, run tests, and iterate on the results. They’re not just answering questions. They’re taking action, autonomously, across your whole codebase.

The word “agentic” comes from “agent,” something that acts on your behalf. In agentic coding, the AI isn’t waiting for you to specify every step. It reads the project, plans its approach, makes coordinated changes, and comes back with a completed feature or fix.

Why It Matters

Agentic coding is the evolution beyond vibe coding. Where vibe coding works great for quick prototypes and single-session experiments, agentic coding is what makes sustained, multi-session projects possible. The AI understands your project as a whole, not just the current prompt.

This is also where the role of the human changes. In vibe coding, you’re the author describing what you want. In agentic coding, you’re more like an architect or product manager: setting goals, defining constraints, reviewing output, and making decisions. The AI handles the implementation.

How It’s Used on This Site

Claude Code operates as an agentic coding tool on this site. When building a new feature, like the glossary system, Claude Code reads the entire codebase, examines the existing patterns, creates new files, modifies existing ones, updates cross-references, and generates content, all in a single coordinated session. The PLAN.md and CLAUDE.md files provide the context that makes this possible: the AI knows the project’s conventions, architecture, and goals before it starts working.