Getting Started with Agentic Coding

Agentic coding is the practice of working alongside AI systems that can take multi-step actions — not just suggest code, but plan, write, test, debug, and iterate autonomously.

What Makes Coding "Agentic"?

Traditional AI coding tools are reactive — you write a prompt, they respond with code. Agentic coding is proactive — you describe a goal, and the AI figures out the steps, executes them, and handles the details.

The key capabilities that enable this:

Tool use — the AI can read and write files, run commands, search the web

Multi-step planning — breaking complex tasks into sub-tasks

Feedback loops — reading test output and fixing failures automatically

Persistent context — remembering decisions across a long session

Setting Up Claude Code

Claude Code is Anthropic's CLI-based agentic coding tool. Install it:

npm install -g @anthropic-ai/claude-code
cd your-project
claude

The CLAUDE.md File

The most important file for agentic coding is CLAUDE.md — a plain-text context file at the root of your project. Claude reads this at the start of every session.

# My Project
Stack

Next.js 14 App Router
PostgreSQL via Prisma
Deployed on Vercel
Commands
npm run dev — start dev server
npm test — run tests
npm run db:push — push schema changes
Code Conventions
Use TypeScript strict mode
Prefer server components; use 'use client' only when needed
All DB queries go in /lib/db.ts

A Practical First Session

Start with a well-scoped task:

"Add a search endpoint to /api/search that queries the products table by name and returns paginated results. Include input validation and error handling."

Claude will:

1. Read your project structure

2. Find the relevant files

3. Write the code

4. Run the linter

5. Report back

Tips for Effective Agentic Workflows

Be specific about outcomes, not steps. Instead of "write a function that does X then Y then Z", say "I need a function that takes user input and returns validated, normalized data — handle edge cases."

Give permission to explore. Tell Claude it can read any file to understand context. The more it understands your codebase, the better its output.

Use /compact for long sessions. When the context gets large, /compact summarizes the conversation and continues with a smaller, focused context.

Commit frequently. Treat every agentic session like a branch. If something goes sideways, git reset is your friend.

Next Steps

Set up your CLAUDE.md today — even 10 lines of context makes a huge difference

Try a refactoring task: "Extract all hardcoded strings in this component into a constants file"

Explore MCP servers to give Claude access to your databases and APIs