Claude Code, Codex and Antigravity: AI agents for building websites and SaaS

AI coding agents can now read a codebase, plan a change, edit files and run tests. As a web developer and AI engineer I use several of them, and each one earns its place for a different kind of work.

What each tool is

• Claude Code: Anthropic's coding agent. It works on your project from the terminal or your editor and can plan, edit and verify changes across many files.
• Codex: OpenAI's coding agent. It takes a task, works on the code and returns changes you can review.
• Antigravity: Google's agent-based development platform, built around agents that plan, write and check software.
• ChatGPT and Claude: general assistants. They are best for thinking, drafting, explaining and comparing options before any code changes.

How I split the work

I use the chat assistants for planning and research: shaping a brief, comparing approaches and writing first drafts. I use coding agents once the plan is clear, for well-defined changes such as a new component, a migration, a bug with a known cause or a round of tests.

Rules I keep with every agent

• Give it a small, specific task and the context it needs, not the whole project at once.
• Work on a branch or a staging copy, never directly on a live site.
• Read every change like a pull request from a new teammate, and run it before it ships.
• Keep secrets, customer data and production credentials out of prompts and out of reach.

Which one should you start with?

Start with the one that fits how you already work: a terminal-first agent if you live in the command line, an editor-based one if you prefer your IDE. The tool matters less than the habits around it. A clear brief, small steps and careful review give good results with any of them.

If you want these tools applied to a real project, see my web development and SaaS projects or get in touch about yours.