# What AI Coding Agents Are Out There

> A practical map of the AI coding tools available today, from IDE extensions to cloud-based agents, and the criteria that actually matter when choosing one.

- **Source:** https://ainativesoftware.engineering/roadmap/day-3/what-ai-coding-agents-are-out-there
- **Site:** AI-Native Software Engineering — https://ainativesoftware.engineering/book

- **Day:** 3 · position 1 of 6
- **Reading time:** 2 minutes
- **Day overview:** [Day 3](https://ainativesoftware.engineering/roadmap/day-3.md)

A practical map of the AI coding tools available today, from IDE extensions to cloud-based agents, and the criteria that actually matter when choosing one.

## Theory

The number of AI coding tools has exploded in the last two years. Trying to follow every launch is exhausting and not particularly useful. What's more useful is understanding the categories and the trade-offs so you can pick the right tool for the right job.

### The main categories

**IDE extensions** Tools that plug into your existing editor (VS Code, JetBrains, etc.) and add AI assistance on top.

**Dedicated IDEs** Editors built from the ground up with AI as a first-class feature. Examples: Cursor, Windsurf, Zed, Kiro (AWS), Google Antigravity.

**Local non-IDE tools** CLI-based agents you run from your terminal, outside any editor. Examples: Claude Code CLI, Codex CLI, Gemini CLI, opencode.

**Cloud agents** Agents that run in remote sandboxes and operate on your codebase autonomously, usually producing a pull request you review later. Examples: GitHub Copilot coding agent, Cursor Background Agents.

### How to evaluate a tool

Instead of picking based on marketing, ask these questions:

- **Model support**: Which models can I use? Can I bring my own API key? Being locked to one model family is a real constraint as the landscape evolves.
- **Pricing model**: Flat subscription, usage-based, or free tier? Usage-based can get expensive fast when running long agent tasks.
- **Open source**: Is the client open source? This matters for auditability and self-hosting.
- **Integration**: Does it fit your existing workflow (GitHub, Jira, Slack)? Friction in integration reduces adoption.
- **Agentic capability**: Can it run multi-step tasks autonomously, or is it primarily autocomplete and chat?

### The landscape changes fast

Tools that are dominant today may be superseded in six months. The skill that doesn't get outdated is knowing how to evaluate and adopt new tools quickly, not memorizing which one is currently on top.

**Theory resources**

- [Coding Agents Comparison: Cursor, Claude Code, GitHub Copilot, and more](https://artificialanalysis.ai/insights/coding-agents-comparison)

## Practice

Run this in a repository you already know, not a toy project.

1. Open the Artificial Analysis coding agents comparison linked above and pick two tools you haven't used before, one IDE extension and one local/cloud agent.
2. For each tool, answer: which models does it support, what is the pricing model, and is the client open source?
3. Look up one tool you already use daily. Find one feature or workflow you haven't tried yet.
4. Write a short personal decision log (3–5 sentences): given your current setup and team, which tool would you try next and why?

- **Previous topic:** [Human in the Loop](https://ainativesoftware.engineering/roadmap/day-2/human-in-the-loop.md)
- **Next topic:** [Getting Started With an AI Coding Assistant](https://ainativesoftware.engineering/roadmap/day-3/getting-started-with-an-ai-coding-assistant.md)

---

_AI-Native Software Engineering by Alfonso Graziano (O'Reilly Media, Early Release; print edition February 2027). Every page of ainativesoftware.engineering is also served as Markdown: append `.md` to any URL. Index: https://ainativesoftware.engineering/llms.txt — whole site in one file: https://ainativesoftware.engineering/llms-full.txt._