Notes6 min read
Claude Code for designers: what it is, and why it isn't just for engineers
Yes. Claude Code is an agent that reads your project, makes a plan, edits files and checks its own work. You describe the outcome and judge the result, so framing, references and critique matter more than knowing how to code.
From the course · lesson 0.1
The short answer
Claude Code is Anthropic's coding agent, and it works well for designers. Its docs describe it as a tool that reads your codebase, edits files, runs commands and works with your other tools. It runs in a terminal, in code editors, in the Claude desktop app and in a browser.
The word that matters is agent. Autocomplete finishes your sentence. An agent takes a goal, looks around, decides on steps, does them, and checks whether they worked. Anthropic's best-practices guide puts it simply: you describe what you want, and Claude explores, plans and builds it.
That changes who can use it. If the work is "describe the outcome, then judge the result," a designer is already trained for it.
What a coding agent does that a chatbot doesn't
A chatbot answers in a window and waits. You copy the answer somewhere, paste it in, and find out later whether it worked. Claude Code works inside your project folder. Three differences follow from that.
It works in your files. It opens the files it needs, so you don't paste context in by hand. Ask "what does this project do?" and it reads the project before answering.
It can check its own work. It can run a build, run tests, or take a screenshot of the page and compare it with your mockup. Anthropic's guide calls a check Claude can run "the difference between a session you watch and one you walk away from." Your job is to say what "done" looks like.
You decide how much it asks first. Claude Code has permission modes, and you switch between them with Shift+Tab:
- Plan mode. Claude reads and writes a plan, but it can't change your files until you approve the plan.
- Manual mode. Claude asks before it edits a file or runs a command.
- Auto mode. A second model reviews each action and blocks the ones that look risky. Recent versions start new terminal sessions in this mode, so check the status bar under the prompt.
Claude Code for designers: directing is a design skill
The skills that make the work good are framing the problem, gathering references, and giving specific critique. You use all three every day.
Anthropic has published how its own teams use Claude Code, and the Product Design team's section is worth reading. The designers there:
- make visual polish changes themselves (typefaces, colors, spacing) instead of writing specs and waiting through rounds of feedback
- paste mockup images into Claude Code and get working prototypes that engineers can build on
- ask Claude to map error states and logic flows, so they find edge cases while designing instead of during development
They also gave one tip that's easy to copy. They wrote a note to Claude saying they were designers with little coding experience who wanted detailed explanations and smaller, step-by-step changes. They said this made Claude's answers much better and the whole thing less intimidating. In Claude Code, that note goes in a file called CLAUDE.md, which Claude reads at the start of every session.
Three things you already know
The course follows one rule: translate, don't replace. Most of what feels foreign has a twin in Figma.
Folders are pages and frames. A project is one folder. Inside it, folders group related things the way pages and frames do. In a project for a made-up testimonial app called Kudos, app/ holds the screens and components/ holds the reusable pieces. Once you know where a screen lives, you can point Claude at it.
Git is version history. Git is the tool that saves named versions of your project, much like Figma's version history. On top of that, Claude Code saves a checkpoint each time you send a prompt. Press Esc twice, or type /rewind, to go back. Checkpoints only cover edits Claude makes with its own file tools, so keep using Git for anything you want to keep.
Skills are components. A skill is a saved set of instructions, written once and reused. It works like a main component: fix it once and every use gets the fix. If you explain the same thing to Claude three times, make it a skill. Your first Claude Code skill takes about fifteen minutes.
Try it: your first hour
- Install. The easiest way in is the Claude desktop app: sign in and open the Code tab. It includes Claude Code and shows changes visually. If you'd rather use the terminal, the docs give a one-line install command and a beginner's terminal guide. The desktop app needs a paid Claude plan; the terminal version also accepts an Anthropic Console account. What Claude Code costs explains the plans and limits.
- Open a project. Use a copy of something small, or a small starter project. Ask:
what does this project do? explain the folder structure. - Introduce yourself. Type
/initto create a starterCLAUDE.md. Then add a line in your own words: you're a designer, you want plain explanations and small changes. - Plan before changing anything. Switch to plan mode and ask for something small and visible, such as:
On the testimonial wall, make the customer's name bolder and add more space between cards. Show me a plan first. - Critique the plan. Is the right file named? Does it touch things you didn't ask for? Say so, then approve.
- Review the result. Look at the page, and look at the diff (the list of lines that changed). Ask "why did you change this file?" about anything that surprises you. If it went wrong, rewind and ask again with a clearer brief.
Where it goes wrong
Generic output. With no direction, AI drifts toward the average. Anthropic's own write-up on front-end design names the defaults it falls back on: Inter and Roboto, purple gradients on white, predictable layouts. The fix is the one you'd use with any new teammate. Write your taste down: references, rules, and things to avoid. Then show it what good looks like.
Fear of the terminal. You need very little of it. Six commands cover most first weeks:
| Command | What it does |
|---|---|
pwd | Shows which folder you're in |
ls | Lists what's in it |
cd folder-name | Moves into a folder |
mkdir name | Makes a new folder |
open . | Opens this folder in Finder (on a Mac) |
claude | Starts Claude Code here |
Inside Claude Code, type / to see every command and skill you can run.
The Foundations stage of Chief of Agents starts here. You set up Claude Code, learn the vocabulary, and write your first skill before you build anything.
Sources
- Claude Code overview, Claude Code docs
- Quickstart, Claude Code docs
- Best practices for Claude Code, Claude Code docs
- Choose a permission mode, Claude Code docs
- Checkpointing, Claude Code docs
- How Claude remembers your project (CLAUDE.md), Claude Code docs
- Extend Claude with skills, Claude Code docs
- How Anthropic teams use Claude Code, Claude blog, and the full report (PDF)
- Improving frontend design through skills, Claude blog
Learn this inside Claude Code.
Join the waitlist for Chief of Agents. You'll get the free beginner's handbook and skills right away, and hear first when the course opens.
Keep reading
7 min read
Agents, skills, MCPs and hooks, explained with Figma
Agent vs skill vs MCP, plus subagents, hooks and plugins: seven Claude Code terms, each matched to something you know in Figma.
6 min read
Your first Claude Code skill in 15 minutes
How to make a Claude Code skill: pick a weekly chore, write it down once as a SKILL.md, and run it any time by typing its name.
7 min read
Connecting Claude to your tools safely: MCP for non-developers
Is it safe to connect Claude to Gmail or Notion? What an MCP server can see and do, why to start read-only, and which hooks help.