Knowing how to prompt is only half the battle. Knowing where and how to input that prompt into your IDE determines your ultimate coding velocity.
1The Chat (The Architect)
The Sidebar Chat should be treated as a senior developer sitting next to you. You do not use it to write 5 lines of CSS. You use it to discuss system architecture, ask for explanations of legacy code, or debug massive stack traces. The Chat window allows you to attach multiple files and maintain a long, conversational context history. However, you must manually move the generated code from the chat into your actual files.
"How should we structure a PostgreSQL database ledger?"
2Inline Edits (The Editor)
Inline editing (Ctrl+K or Cmd+K) is surgical. It happens directly inside your active code file. You highlight a function, press the shortcut, and give a brief command like 'Optimize this query'. The AI does not talk back; it instantly generates a code Diff (Red for old code, Green for new code). You review the Diff and press Enter to accept. This is the fastest way to refactor existing logic.
// Instant diff view updated inline
3Autocomplete (The Typist)
Ghost Text (Autocomplete) is completely passive. You do not prompt it. As you type, the AI constantly analyzes your current file and attempts to predict your next keystroke. It is incredibly useful for repetitive tasks, such as filling out large arrays of data, mapping over objects, or writing standard error handlers. Simply press 'Tab' to accept the AI's prediction.
// Ghost text: age: 24, email: "[email protected]"
