AI coding tools differ mainly in how much of the work they take on: from answering one question, to suggesting the next line, to editing a whole repository. Knowing the four kinds helps you pick the right one for a task.
The four kinds#
| Type | How you use it | Good for |
|---|---|---|
| Chat assistant | Describe a problem in plain language and get code, explanations or debugging help back. | Exploring unfamiliar APIs, understanding error messages, generating boilerplate, thinking through design decisions |
| Autocomplete | Inline suggestions in your editor as you type, predicted from the current file and context. | Repetitive patterns, obvious implementations, less typing |
| IDE agent | Describe a feature or refactor; the tool edits several files, runs commands and iterates. | Larger, scoped tasks inside a codebase |
| Command-line agent | A terminal tool that reads and writes files, runs commands and works across the whole repository. | Longer tasks: a feature end to end, running tests, fixing a class of bugs across many files |
More scope, more to review#
The further right a tool sits on that scale, the more it does without you watching each step. That saves time, but it also means a bigger change to check afterwards. The workflow note covers how to keep that change reviewable.
Short line#
Chat assistants answer questions and explain code, autocomplete fills in the next line, IDE agents make multi-file changes toward a goal, and command-line agents work across a whole repository. More scope saves more time but gives you more to review.
Key takeaways#
- Chat assistant: questions, explanations, debugging, boilerplate, design thinking.
- Autocomplete: inline suggestions for repetitive or obvious code.
- IDE agent: multi-file edits toward a goal, with commands and iteration.
- Command-line agent: repository-wide, longer tasks such as a feature end to end.
- The more a tool does on its own, the larger the change you must review.