Getting Started with Agentic Coding

Agentic coding is the practice of using AI agents — like Claude Code, Codex CLI, or Cursor — as active collaborators in the development process, rather than just autocomplete tools. This guide covers what it is, how it differs from traditional AI coding, and how to start effectively.

What Makes It “Agentic”?

The key difference from traditional AI-assisted coding:

	Traditional AI Assist	Agentic Coding
Scope	Single lines / functions	Entire features across files
Interaction	You type, it autocompletes	You describe intent, it plans and executes
Context	Current file only	Reads your codebase, project rules, docs
Memory	None between prompts	Session context, CLAUDE.md, memory files
Decision-making	You drive everything	Agent makes decisions, you review
Tool use	Suggestions only	Reads files, runs commands, creates PRs

The shift is from “smarter autocomplete” to “junior developer that works fast, reads everything, and needs code review.”

What Agents Are Good At

Based on real experience building krowdev and WebTerminal:

• Reading large codebases fast — an agent analyzed 11 terminal emulator source repos in hours, extracting architecture patterns that would take weeks manually
• Consistent formatting and boilerplate — schema definitions, test scaffolds, CSS custom properties
• Cross-file refactors — renaming a concept across 15 files, updating imports, fixing references
• Research synthesis — reading docs, comparing approaches, summarizing trade-offs
• Mechanical work you understand — “add breadcrumbs to every entry page” when you know exactly what breadcrumbs should look like

What Agents Struggle With

• Taste and judgment — they’ll over-engineer, add unnecessary abstractions, and optimize things that don’t need optimizing
• Knowing when to stop — without constraints, they’ll keep “improving” code until it’s unrecognizable
• Your project’s history — they don’t know why a decision was made, only what the code looks like now
• Novel architecture — they recombine patterns from training data, they don’t invent genuinely new approaches
• Subtle bugs — they’re confident, not careful. Their code works on the happy path but may miss edge cases

Core Patterns

This knowledge base documents the patterns that make agentic coding work:

Pattern	What it solves	Guide
Prompt Patterns	Getting better results from each interaction	Read →
Context Management	Feeding agents the right information	Read →
Codebase Research	Systematically analyzing existing code	Read →

Start Here

Your first agentic task should be small, well-defined, and reviewable:

1. Pick a task you already know how to do — so you can evaluate the agent’s output. A bug fix, a utility function, a styling change.
2. Write a clear prompt describing the what and why, not the how. “Add a 404 page that matches the site design with links back to the homepage and explore page” is better than “create src/pages/404.astro with an h1 and two anchor tags.”
3. Let the agent propose before it builds. If you’re using plan mode or asking for an approach first, you catch bad ideas before they become bad code.
4. Review the output like a code review. Read every changed line. Agents are confident — they’ll commit to an approach even when it’s wrong. Your job is to catch the 10% that’s subtly incorrect.
5. Document what you learn. The prompt that worked, the constraint that prevented over-engineering, the anti-pattern that wasted an hour. That’s what this knowledge base is for.

Your second task should use CLAUDE.md. Create a project rules file before starting. Even 10 lines of stack + conventions context dramatically improves output quality. See Context Management for patterns.