Lesson 1 / 25

Autocomplete, Chat and Agents

Tell apart the three kinds of coding assistant and know what each is good for.

Three modes of help

Inline completion suggests the next line or block as you type, best for boilerplate and familiar patterns. Chat answers questions, explains code and drafts snippets you copy in. Agents go further: they read many files, edit them, run tests and commands, and iterate until the task passes. Most tools (GitHub Copilot, Cursor, Claude Code, JetBrains AI and others) now offer more than one mode, and their features change often.

From suggestion to action

Assistants range from autocomplete to chat to agents that edit files and run commands. More autonomy means more review.

Four levels: complete, chat, edit, act.
Figure 1.1 — Complete, chat, edit and act.

Matching the mode to the job

Use the lightest mode that does the job. Higher autonomy saves typing but needs more careful review of the result.

Finish this line / repeat a pattern     -> inline completion
"What does this regex do?"                 -> chat
"Rename this function everywhere"          -> agent / multi-file edit
"Fix the failing tests in this module"     -> agent that can run tests

Spell-check, tutor, intern

Autocomplete is like spell-check: quick and local. Chat is a tutor you can ask. An agent is a junior intern who can do a whole task but whose work you must review before it ships.

Quick check: Which kind of assistant can edit many files and run your tests?

  • Inline completion only
  • A spell-checker
  • An agent mode
  • A syntax highlighter
Answer

An agent mode — Agents combine file access, editing and command execution in a loop.