AI Tools Overview

Two categories of AI coding tools

AI coding tools fall into two broad categories, and understanding the difference is the most important thing in this guide.

Hosted builders — tools where you build entirely in the browser. You don't install anything. You don't see the code files on your computer. Everything lives on their servers. Examples: Replit, Lovable, Bolt.

Local-first tools — tools where you work on your own computer, in your own IDE, with your own files. AI helps you write code, but the code lives on your machine. Examples: Cursor, Claude Code, GitHub Copilot, Codex.

This distinction matters more than almost any other choice you'll make.

Hosted builders: where most people start

Replit

Replit is a browser-based development environment with AI built in. You open a browser tab, describe what you want, and it generates a working app. You can see and edit the code, run it, and even deploy it — all without leaving the browser.

Good for: Quick prototypes, learning, experimenting with ideas. Zero setup required.

Limitations: Your code lives on Replit's servers. You're dependent on their platform. Performance can be limited. As your project grows, you'll hit walls — limited file structure control, harder debugging, and less flexibility with databases, APIs, and deployment options.

Lovable

Lovable focuses on building web apps through conversation. You describe what you want and it generates a polished UI with working functionality. It's particularly good at producing attractive front-end designs quickly.

Good for: Rapid prototyping, MVPs, getting a visual concept in front of people fast.

Limitations: Similar to Replit — your code lives on their platform. Customization gets harder as complexity grows. Moving your project off Lovable and onto your own infrastructure requires significant rework.

Bolt

Bolt (by StackBlitz) is another browser-based AI builder. You describe what you want and it generates full-stack applications. It runs entirely in the browser using WebContainers technology.

Good for: Quick prototypes, trying out ideas, generating starter code.

Limitations: Same category of trade-offs — browser-based, platform-dependent, harder to scale beyond prototypes.

Local-first tools: where real building happens

These tools run on your computer. Your code lives on your machine. You have full control over your project structure, your database, your deployment, everything. The AI is a powerful assistant, but you own the code.

Cursor

Cursor is a full IDE (code editor) with AI deeply integrated. It's built on VS Code, so it looks and works like the most popular code editor in the world, but with AI woven into every part of the experience.

What it does: AI chat panel for asking questions and generating code. Inline editing — highlight code and ask AI to change it. Tab completion that predicts what you're about to type. Agent mode that can make changes across multiple files. Supports multiple AI models (Claude, GPT, Gemini) — you pick which one to use.

Best for: Your primary development environment. This is where you spend most of your time building. If you're following our guides, this is what we recommend as your IDE.

Cost: Free tier available. Pro plan starts at $20/month with a credit pool for using premium AI models.

We cover Cursor setup in detail in the IDE Setup guide.

Claude Code

Claude Code is Anthropic's command-line coding tool. Instead of a visual editor, you interact with it through your terminal. You describe what you want, and it writes code, creates files, runs commands, and makes changes to your project — all from the command line.

What it does: Understands your entire project (reads your files, understands how they connect). Can create, edit, and delete files. Can run terminal commands. Works as an autonomous agent — you describe a task and it figures out the steps.

Best for: Larger tasks where you want AI to handle multiple steps autonomously. Setting up project scaffolding. Making changes that span many files. Developers who are comfortable with the terminal.

Cost: Requires an Anthropic API key. You pay per usage based on the tokens consumed.

GitHub Copilot

GitHub Copilot is an AI coding assistant that plugs into VS Code (and other editors). It's best known for its autocomplete — it watches what you're typing and suggests the next lines of code in real time.

What it does: Real-time code suggestions as you type. Chat panel for asking questions. Code explanation and generation. Integrates with GitHub for pull request summaries and code reviews.

Best for: Developers who want AI assistance inside VS Code without switching to a different editor. The autocomplete is excellent for writing code faster.

Cost: Free tier available for individual use. Pro plan starts at $10/month.

Codex (OpenAI)

Codex is OpenAI's coding agent, available through ChatGPT. It can read your codebase, make changes, and even run tests. It works as a cloud-based agent — you give it a task, it works on it, and you review the results.

What it does: Reads and modifies code in a GitHub repository. Runs in the cloud — works even when your computer is off. Can handle multi-file changes, write tests, fix bugs.

Best for: Delegating coding tasks you don't need to watch in real time. Bug fixes, refactoring, writing tests, documentation.

Cost: Included with ChatGPT Pro ($20/month) with usage limits. Higher tiers available.

Which should I use?

Here's the simple version:

If you're just starting out and want to see what AI coding feels like, try a hosted builder like Replit or Lovable. Zero setup, instant results. Great for your first few experiments.

When you're ready to build something real, move to a local-first setup. Install Cursor (or VS Code with an AI extension) as your IDE. This is where you'll do your actual development work.

As you get more comfortable, add tools as needed. Claude Code for bigger autonomous tasks. GitHub Copilot if you prefer VS Code. Codex for tasks you want to delegate.

The landscape changes fast

New AI coding tools launch constantly. Models get smarter. Features get added. Pricing changes. What's true today might shift in six months.

That's okay. The core pattern stays the same: you describe what you want, AI helps you build it, and the code lives somewhere (their servers or your computer). The concepts from the book — architecture, state management, debugging — apply no matter which tool you're using.

When in doubt, choose tools that give you ownership of your code (local-first), flexibility to switch models (not locked into one AI provider), and access to a real development environment (terminal, file system, databases, deployment options). Those principles will serve you well even as specific tools come and go.