# Working with AI agents

Open the project in Claude Code, Cursor, Codex, Copilot, or Zed. Your agent reads `AGENTS.md` and the skill files, then builds inside the structure that's already there, so folders, naming, and styling stay the same across sessions.

You can point an agent at an empty repo and ask for an admin dashboard, and it will get you something. The problem shows up on the second and third session, when it picks a different folder layout, a different auth approach, and styles components a little differently each time. You end up fixing patterns instead of shipping features.

NextAdmin gives it a starting point that doesn't move. The components, the auth screens, the data shape, and the rules are already decided, so each session picks up where the last one left off.

A few things worth knowing:

- **Agent Skills** give your agent the full picture of the components, design system, and file structure, so it doesn't start from zero
- **React Aria skills** keep keyboard navigation, focus, and ARIA attributes correct after your agent edits a component
- **The api-integration skill** swaps the mock data for your live endpoints from a single prompt
- **Custom Prompts** cover the common jobs across core setup, layout, UI, and auth, ready to copy into your editor