The AI Coding Tools Landscape: 10 Categories to Help You Pick the Right One

Google's CEO revealed in April that 75% of Google's new code is now generated by AI, then reviewed by engineers. Behind that number lies an AI coding tool ecosystem that's exploding into dozens of distinct categories — yet most people only know one or two of them. This article breaks down today's AI coding tools into ten categories, helping you understand what each one solves, who it's for, and what the real costs and risks look like.

Chat Assistants: Where Almost Everyone Starts

ChatGPT, Claude, and Gemini are the first stop for most people getting into AI-assisted coding. Paste in some code or an error message, ask "what's wrong here?", and they're great at explaining code you don't understand, fixing bugs one at a time, or writing small scripts.

The limitations are obvious: they can't see the rest of your project, so you're stuck copying and pasting back and forth. There's also a trust issue worth noting — in the Stack Overflow 2025 Developer Survey, more developers distrusted AI tool accuracy (46%) than trusted it (33%). No matter how convenient the tool, running the code before relying on it is always the baseline.

Autocomplete: Smallest Footprint, Fastest Results

In-editor autocomplete is best known through GitHub Copilot. As you type, it predicts the next line — sometimes a whole function — and a Tab keystroke accepts the suggestion. The free tier offers 2,000 completions and 50 chats per month; Pro is $10/month for unlimited completions.

The $39 and $100 tiers are really for people running agents continuously — both chat and agent work now draw from a shared monthly AI credit pool. If you already know how to code, this is the best starting point: it changes your workflow the least and delivers value from day one.

AI Code Editors: IDEs Rebuilt Around AI

Cursor, Devin Desktop (formerly Windsurf), Google Antigravity, and Amazon's Kiro are full editors built from the ground up around AI. Instead of completing a single line, they let AI plan and edit multiple files simultaneously while you review the changes.

Cursor has the highest name recognition, with a free tier and a Pro plan at $20/month. Devin Desktop is the new name for Windsurf, now developed by Cognition — the company behind the Devin agent. Existing Windsurf users don't need to do anything; it'll arrive as a regular update with plans, extensions, and settings all migrated over. Google Antigravity is its agent-first platform, free for individual developers. Kiro is AWS's editor, with a free tier offering 50 credits.

Terminal Agents: Fully Autonomous Executors in the Command Line

Claude Code, OpenAI's Codex, Gemini CLI, and OpenCode live in the terminal. You describe a task in natural language, and the agent reads your project, edits files, runs commands and tests, and keeps going until the job is done.

Claude Code is available with Claude's $20/month Pro plan and Max plans; Codex is included in all ChatGPT plans — even the free tier — and can run in both the editor and the web interface. Gemini CLI is open source, with up to 1,000 free daily requests for personal Google accounts. OpenCode is also open source and connects to over 75 model providers, including local models. One warning: these agents can execute commands on your machine. Always start them on projects with backups, and review every change before accepting it.

The core mechanism behind terminal agents is Tool Use / Function Calling: rather than just generating text, the model can invoke predefined tools — reading files, writing files, executing shell commands, running test suites. After each step, the tool's output is fed back to the model as new context, which then decides the next action. This forms an autonomous agentic loop that runs until the goal is complete or an unresolvable error is hit. The key difference from a regular chat assistant: it's not just giving suggestions — it's actually operating on your filesystem. That's precisely why running an agent directly on an unfamiliar codebase or production environment carries far greater risk than executing commands manually.

Cloud Agents: Hand Off the Work and Close Your Laptop

Codex's web interface, Google's Jules, and Devin run on remote servers. You hand off a task — fix a bug, upgrade a dependency — and you can close your laptop. The agent works in its own copy of the project, often handling multiple tasks simultaneously, and when it's done you review the results the same way you'd review a colleague's work. Jules starts at 15 tasks per day.

These tools are best suited for tedious, well-defined work: writing tests, bumping version numbers, small bug fixes. With vague requirements, you'll spend more time explaining than you would have spent writing the code yourself.

App Builders: The Entry Point for Non-Coders

Lovable, Bolt, Replit, and V0 are aimed at users with zero coding background. You describe the app you want, it builds a clickable, working version, and you keep asking for changes — this is what people call vibe coding.

Lovable gives 5 free daily build credits; Bolt's free tier includes 1 million tokens per month, with a Pro plan at $25; Replit starts at $20/month and handles both building and hosting; Vercel's V0 offers 7 free messages per day on the free tier. These are great for prototypes and personal tools — but before real users sign up, make sure someone checks authentication and data handling. Security is the part you can't see.

"Vibe Coding" is a term coined by OpenAI co-founder Andrej Karpathy in early 2025 to describe a style of programming driven entirely by natural language, with little to no direct interaction with actual code. The creator steers AI iteration by continuously describing desired outcomes rather than writing logic line by line. This dramatically lowers the barrier to software creation — but introduces a systemic risk: the generated code is a black box to its creator. Problems are hard to diagnose, and security vulnerabilities (missing authentication, SQL injection, exposed sensitive data) are especially easy to miss. App builders are great for rapid idea validation, but bringing in an experienced developer for a security review before opening access to real users is a necessary step, not an optional one.

Design-to-Code: The Natural Entry Point for Designers

Figma Make and V0's design mode start from "what should this look like." Give it a design mockup or a text prompt and it produces a working prototype backed by real code. Figma Make lets you switch between code and canvas and continue editing visually. For designers, this is the most natural onboarding because you stay inside familiar tools.

AI Code Review: Review Is Becoming the New Work

CodeRabbit and GitHub Copilot's code review feature read your changes when you open a Pull Request and leave comments — like a tireless senior engineer. CodeRabbit is free forever for public repositories, with paid plans starting at $24 per developer; GitHub Copilot can review PRs directly inside GitHub on paid plans.

Circling back to that Google 75% — those commits are reviewed by human engineers. As AI writes more and more code, reviewing that code is becoming the core of the job.

A Pull Request (PR) is the central workflow of modern software collaboration: a developer submits their changes on a separate branch and opens a merge request; team members review before the changes land in the main codebase. AI code review tools hook into platforms like GitHub and GitLab via webhooks, automatically fetching the diff when a PR is created or updated, analyzing change logic, potential bugs, security risks, and code style, then surfacing results as inline comments — exactly the same interface human reviewers use. The value of these tools is catching details that humans miss under time pressure or fatigue, especially as AI-generated code floods codebases and automated review becomes a critical quality-control layer.

Open-Source Extensions: Bring Your Own API Key, Pay Per Use

Open-source extensions like Cline plug into VS Code and work like an agent — but are completely free and open source. Cline is Apache 2.0 licensed with nearly 70,000 stars on GitHub. You bring your own API key from any major provider, or point it at a local model, paying only for what you use and switching models whenever you like.

One note: Cline's popular fork RooCode shut down on May 15, 2026; its repository now redirects users to Cline and a community fork.

Local Free Tools: Code That Never Leaves Your Machine

Ollama paired with open-source coding models like Qwen Coder lets you download a model once and run it entirely locally. Your code never leaves your machine — which is critical for many companies. Qwen's coding models range from 0.5B to 32B parameters, with smaller sizes running fine on a regular laptop. Connect them to Cline or OpenCode and you have a free coding agent.

To be honest, local models are still slower and less capable than large cloud models — but for sensitive codebases, or for anyone who wants to learn at zero cost, they're hard to beat.

Ollama is a local model runtime that wraps the complex process of downloading, managing, and serving open-source LLMs into a simple CLI tool. It starts a local server endpoint compatible with the OpenAI API format (default: localhost:11434), meaning virtually any AI coding tool that supports a custom API address — Cline, OpenCode, Continue, and others — can connect to local models with no code changes required. The Qwen Coder series is open-sourced by Alibaba Cloud and fine-tuned specifically for coding tasks, with standout performance among open-source models at comparable parameter counts. The 7B and 14B versions run smoothly on a regular laptop with 16GB of RAM, making them one of the best-value choices for local coding model deployment today.

Closing Thoughts: Where Should You Start?

• Never coded before, or still learning: Start with an app builder, then layer in a chat assistant and GitHub Copilot's free tier.
• Already a developer: Pick one AI editor or one terminal agent, and go deep on just that one.

The number of tools is exploding, but the decision logic stays the same — find the category that best matches where you are right now, then get good at it.