AI Guides › Step-by-step guides
By Nigel Guy · 6 min read
Most people open Claude, type a question, read the answer and copy it somewhere else by hand. It feels fine because the answers are good. What it hides is that you are doing the doing: the pasting, the file shuffling, the running of code. Claude has three ways of working, and the choice between them is really a choice about who handles the output.
The rule: pick the mode by what you want at the end, not by how hard the task feels. A reply to read means Chat. Finished files on your computer mean Cowork. Working software in a project means Code.
You need a Claude account and, for two of the three modes, a paid plan. Here is how it stands at time of writing, from Anthropic's pricing and help pages:
| Mode | Where you find it | Plan needed |
|---|---|---|
| Chat | claude.ai, desktop app, mobile app | Free works |
| Cowork | Claude Desktop (macOS and Windows), plus web and mobile on Pro, Max and Team | Paid plan, Pro at minimum |
| Code | Terminal, VS Code, JetBrains, desktop app Code tab, claude.ai/code | Paid subscription or an Anthropic Console account for most surfaces |
Pro is listed at $20 a month, or $17 a month billed annually (about £15 and £13 at time of writing). Max starts from $100 a month (about £75 at time of writing). Check the £ price at checkout, because exchange rates and tax move it. Pricing pages change often, so treat these as a snapshot.
You also need one real job. Not "explore Claude": a specific task you would otherwise do by hand this week.
Finish this sentence: "When this is done I want [OUTPUT]." Be literal. "An answer I can read" is an output. "A tidy folder of renamed invoices" is an output. "A script that renames them and a test that proves it" is an output. If you cannot name it, you are still thinking, and Chat is the right place to do that.
| If the output is... | Use | Why |
|---|---|---|
| An answer, draft, plan, explanation or critique you will read and reuse | Chat | One conversation, you stay in control of every step |
| Files, spreadsheets, summaries or organised folders from many small steps | Cowork | Anthropic describes it as bringing Claude Code's agentic abilities to knowledge work beyond coding: it plans, works through steps and delivers finished files |
| Code that is written, run, tested and committed in a project | Code | It reads your codebase, edits files, runs commands and works with git |
A fast tie-breaker: if you would be content with Claude telling you what to do, use Chat. If you want Claude to do it to files, use Cowork. If those files are a codebase, use Code.
Chat is the default, so most mistakes here are about under-briefing. Open claude.ai or the app, start a new conversation and give it four things: who you are, what the job is, what a good answer looks like, and the material to work from. Chat suits drafting, thinking through a decision, studying a document you paste in, and anything where you want to steer each reply. The free plan includes chat, web search, file creation, code execution, memory and artifacts at time of writing.
Move on from Chat when you notice you are repeating the same copy-paste routine across many files. That is the signal for Cowork.
Cowork is for multi-step work on files: sorting a messy downloads folder, turning receipts into a spreadsheet, pulling notes from a pile of documents into one summary. Per Anthropic's help centre, it can continue a task without you watching each step, and it has deletion safeguards that ask for explicit approval before removing files.
Multi-step tasks use noticeably more of your usage allowance than a simple question, so expect them to cost more than a Chat reply.
Code is for software. Easiest route if you are not technical: install the Claude desktop app and use its Code tab, which includes Claude Code, so no separate install is needed. From a terminal, Anthropic's current native install on macOS, Linux and WSL is:
curl -fsSL https://claude.ai/install.sh | bash
On Windows PowerShell it is irm https://claude.ai/install.ps1 | iex. Open a new terminal and run claude --version to confirm it worked, then cd into a project folder and run claude. It will ask you to log in on first use. Ask for one small, checkable change first, such as "add a test for this function and run it".
If you have no codebase and no wish to build one, you do not need this mode.