Claude, Cursor, Codex, Antigravity, and company

The ecosystem changes fast, but the categories don't. Any new tool that comes out falls into one of the three modes from the last lesson (chat, AI-powered IDE, agent) and uses one of the models you'll see in the next two lessons. Here's a map of the ones that matter today and how I use them, without making your head spin.

A clarification that confuses everyone at first: there's a difference between the MODEL (the brain: Claude, GPT, Gemini) and the TOOL (the app where you use it: the Claude chat, Cursor, Codex). The same tool can use different models under the hood. Cursor, for example, lets you choose which model it works with.

Claude (Anthropic)

My main model for building. It's excellent at code, at carefully following instructions, and at long-form reasoning without getting lost. It exists as a chat (claude.ai) and as an agent (Claude Code) that works inside your project, running tasks in the terminal. When something is delicate or long, it's my go-to.

Cursor

A code editor (based on VS Code, so if you already know VS Code, you'll feel right at home) with very powerful built-in AI: it autocompletes, edits multiple files at once, and understands your entire project. It's the most comfortable entry point to the 'AI-powered IDE,' and it lets you choose which model to use under the hood (Claude, GPT, Gemini).

Codex and code agents

Codex (from OpenAI) and similar agents run programming tasks autonomously. You describe what you want, and they work on their own through several steps: reading, writing, testing. They're very powerful when you already know how to direct and review their work — less recommended as a first step, because they do a lot at once.

Antigravity and the newcomers

New tools come out every month (Google's Antigravity is one of them). Don't stress about trying them all: almost all of them are just a variation of an AI-powered IDE or an agent. Learn the CONCEPT of the three modes, and any new tool will take you minutes to understand — just ask yourself, 'Is this a chat, an IDE, or an agent?' and you'll know how to use it.

A practical example: what I grab for each task

On a normal day of building, my choice is almost automatic:

• Thinking through an idea or understanding a new concept → Claude's chat. Total control, zero risk of it touching my code.
• Quick edits while I'm writing, tweaking a component I have open → Cursor. The AI sees the file and goes at my pace.
• A large, repetitive task (renaming something in 20 files, migrating a library) → an agent (Claude Code). I give it the instruction, review the diff, and approve.

Your takeaways

• Model (the brain) and tool (the app) are different things; one tool can use multiple models.
• Every new tool falls into one of the three modes — identify it, and you'll know how to use it.
• My combo: Claude (chat + agent) + Cursor. Yours can be just a chat at the beginning.
• Master a few tools deeply instead of trying many superficially.

The best tool isn't the newest one: it's the one you've mastered well enough to build with it today.