Most people start out using Claude like a chat app: question in, answer out. In Claude Code, though, Claude becomes an agent that does the work itself. Here are three concepts worth knowing.

1. The agent: a loop that gets the job done

An agent takes a task and works through it in a loop: it gathers context, takes action and verifies the result, repeating until the job is done. Tools are what make it an agent. It can read files, edit code, run commands and search the web, and you can step in and change direction at any time.

Tip: describe the goal and the context as if you were handing the task to a colleague, rather than listing every step.

2. Subagents: delegation with a clean slate

A subagent is a specialised assistant the main agent hands a well-defined piece of work. Each subagent runs in its own context window, with its own instructions, tools and permissions, and returns only a summary. Your main conversation doesn’t fill up with search results and intermediate steps.

Claude Code ships with built-in subagents, such as Explore, which searches code without changing anything. You can also create your own as a Markdown file in .claude/agents/ for one project, or ~/.claude/agents/ for all your projects:

---
name: reviewer
description: Reviews code changes and points out bugs
tools: Read, Glob, Grep
---

You are a code reviewer. Go through the changes and give clear, actionable feedback.

Subagents work well for searching a large codebase, code review, and tasks that can run in parallel.

3. Scheduled tasks: Claude works while you do something else

There are three ways to have Claude work on a schedule:

  • /loop repeats a prompt while the session is open, for example /loop 5m check whether the deploy finished. Recurring tasks expire automatically after seven days.
  • Scheduled tasks in Claude Desktop run on your computer, for example daily or weekly, with access to your files. The computer has to be awake and the app open.
  • Routines run in the cloud, even when your computer is off, on a schedule, on an API call or on GitHub events. You create them at claude.ai/code/routines or with /schedule.

Rule of thumb: /loop to keep an eye on something right now, Desktop when the task needs your files, and routines when it should run on its own.

References