Autocomplete vs. agent
The fastest way to understand Claude Code is to contrast it with the AI tooling you already know. A code-completion engine sees the file in front of you and predicts the next few tokens. It is fast and narrow, and you remain responsible for every decision and every keystroke.
Claude Code operates at a different altitude. You hand it a task — "add input validation to the registration form" — and it reads the relevant files, plans an approach spanning several of them, makes the edits, runs the tests, reads the failures, and iterates until the task is actually done. That is not autocomplete. It's delegation.
Code completion
Sees one file, suggests the next line. You drive every keystroke. Great for speed, blind to the rest of the project.
Claude Code (agent)
Reads the whole project, plans across files, executes, runs tests, and fixes failures on its own. You give direction, not keystrokes.
The agentic loop
Everything Claude Code does runs through one repeating loop. Each phase chains naturally into the next, and Claude can run dozens of tool calls inside a single turn to finish your task.
Gather context
Read source files, search the codebase with grep and glob, check git status and recent history, and load CLAUDE.md instructions.
Take action
Edit files, create new ones, run shell commands, build, and make commits — within whatever permission mode you've set.
Verify results
Run tests, read output, compare against the expected behaviour, and loop back to step 1 to fix anything that's wrong.
Why it matters: Because Claude verifies its own work, a failing test isn't the end of the turn — it's just another signal to gather context and try again. We cover this loop hands-on in the next two chapters.
The five tool categories
Without tools, a language model can only produce text. Tools are what let Claude Code act. The built-in tools fall into five families:
File operations
Read, edit, create, and rename files anywhere in the project — not just the one you have open.
Search
Pattern matching, regex, and codebase exploration to locate functions, references, and entry points fast.
Execution
Run shell commands, test suites, git operations, and build tools — then read the output.
Web
Search the web and fetch documentation when an answer lives outside your repository.
Code intelligence
Type errors, go-to-definition, and find-references via language-server plugins (LSP) — for semantic understanding beyond plain text search.
Choosing your surface
Claude Code runs the same agentic engine everywhere — the surface only changes your interface, not the model's capabilities. Pick the one that matches the task in front of you.
| Surface | Best for | Standout strength |
|---|---|---|
| Terminal CLI | Power users, automation, scripting | Full power, composable with pipes and scripts |
| VS Code extension | Staying inside your editor | Inline diffs and @-mentions |
| JetBrains plugin | IntelliJ, PyCharm, WebStorm users | Same in-editor flow for JetBrains IDEs |
| Desktop app | Reviewing changes, running several sessions | Visual diff review, multiple sessions, scheduled tasks |
| Web | Zero local setup, cloud-run work | No install, mobile access |
| iOS / Slack | On the go, team workflows | Kick off and check on work from your phone or chat |
There are also three execution environments behind these surfaces — local (runs on your machine), cloud (runs on Anthropic's infrastructure), and remote control (you steer a session running elsewhere). We return to these when we cover automation and routines.
Common mistakes to avoid
- Treating it like a chatbot. Claude Code is a development partner you delegate tasks to, not a Q&A box you query.
- Expecting outside knowledge. It works with what it can see in your project — give it the context it needs.
- Confusing it with Claude.ai. Different products with different purposes; Claude Code is built for engineering work.
Hands-on exercise
List five real development tasks you do every week. For each one, identify which of the five tool categories Claude Code would use — file operations, search, execution, web, or code intelligence — and which surface (terminal, IDE extension, Desktop app, web, or Slack) would feel most natural. The goal is to start thinking in terms of delegation rather than autocomplete.
Further reading
Go straight to the source to dig deeper into anything on this page:
- code.claude.com/docs/en/overview — the official Claude Code documentation and overview.
- anthropic.com/product/claude-code — the product page with capabilities and surfaces.
Claude Code Basics FAQ
What is Claude Code in simple terms?
Claude Code is an agent you delegate whole tasks to, not a chatbot you query. You give it a task and it reads the relevant files, plans an approach across them, makes the edits, runs the tests, and iterates until the task is done.
How is Claude Code different from autocomplete?
Autocomplete sees the file in front of you and predicts the next few tokens. Claude Code reads the whole project, plans across files, executes changes, runs tests, and fixes failures on its own. That is delegation, not line completion.
What is the agentic loop?
The agentic loop is the repeating cycle behind everything Claude Code does: gather context, take action, then verify results. A failing test is not the end of the turn, just another signal to gather context and try again.
What tools can Claude Code use?
Claude Code's built-in tools fall into five families: file operations, search, execution of shell commands and tests, web search and fetch, and code intelligence such as type errors and go-to-definition via language servers.
Which Claude Code surface should I use?
The same agentic engine runs everywhere, so pick by task: the terminal CLI for power and scripting, the VS Code or JetBrains extensions to stay in your editor, the desktop app for reviewing changes, and web or Slack for working on the go.
Quick summary
- Claude Code is an agent that delegates whole tasks — not a line-completion tool
- It runs one loop: gather context → take action → verify results
- Five tool categories let it act: file ops, search, execution, web, code intelligence
- The same engine powers every surface — terminal, VS Code, JetBrains, Desktop, Web, and Slack