Your conversational AI — claude.ai in a browser or the desktop app.
The mental model
Think of Claude Chat as a brilliant new coworker who just got onboarded to your project. They're
sharp, capable, and ready to help with almost anything — but they don't have the project history
you do. They don't know the client's quirks, the decisions that got reversed two months ago, or
what the brand guide actually means in practice. You have to bring them up to speed at the start
of every conversation.
That's the default state of Claude Chat: highly capable, but starting fresh each time. Each new
conversation is a clean slate — Claude has no memory of anything you discussed in a previous chat
unless you bring it back up yourself, or unless you've enabled Memory (more on that below).
That said, within a single conversation, Claude remembers everything — every message, every file
you've shared, every decision you've talked through. It's only when you start a new conversation
that the slate gets wiped.
Memory — turning that new hire into a real colleague
There's an opt-in feature called Memory that changes things meaningfully. When it's on, Claude
pays attention across your conversations and builds up a running picture of who you are, how you
work, and what matters to you — your preferences, your team's conventions, recurring client context,
and so on. That context then gets quietly carried into every new conversation.
With Memory off, Claude is that brilliant new hire: capable, but needs briefing every time.
With Memory on, Claude is more like a real colleague — someone who remembers that you had this
conversation last week, knows your preferred tone for that client, and doesn't need you to explain
your workflow from scratch.
A few things worth knowing:
🔍
You're in control of what it remembers
Go to Settings → Capabilities → "View and edit your memory" to see exactly what Claude has noted down. You can edit or delete anything.
🗂️
Each Project has its own separate memory
Memory inside a Project stays scoped to that Project — it doesn't bleed into your general conversations or other Projects.
👻
Incognito mode skips memory entirely
Click the ghost icon in the top-right corner of a new chat for a conversation that isn't saved to memory or chat history — good for sensitive brainstorms.
💬
You can update it mid-conversation
Just tell Claude "remember that…" and it'll update immediately, without waiting for the daily refresh.
Memory is off by default. Enable it in Settings → Capabilities whenever you're ready to try it.
What designers use it for
✍️
Writing and editing
UX copy, microcopy, design doc drafts, presentation narratives, design critiques, client emails.
🧠
Brainstorming and research
Exploring concepts, competitive research, getting a second opinion on a design direction, mapping user flows in conversation.
📄
Working with documents
Upload a brief, a spec, a transcript, or a PDF and ask Claude to summarize, find gaps, pull out key insights, or draft a response.
🎨
Generating artifacts
Claude can produce things like HTML mockups, SVG illustrations, or styled documents right in the chat — viewable in a side panel.
A dedicated workspace for ongoing work — with shared context built in.
The mental model
A Project is still Claude Chat — same interface, same conversation style — but now that brilliant
coworker has been on the project with you since the beginning. You've given them a proper brief
pack: the brand guidelines, the client history, the tone of voice doc, the decisions that were
already made. Every time you sit down together, they're fully read-in. You don't re-explain; you
just get to work.
You set that context up once — by uploading reference files and writing a set of instructions —
and it's available in every conversation you have inside that Project, from that point on.
There's one important constraint to understand, though. Whatever Claude produces in a
conversation — a draft, a revised brief, a set of copy options — it hands back to you like a file
over Slack. You still have to take that output and do something with it yourself: download it,
upload it to Box, paste it into your doc, integrate it into your design. Claude can't reach into
your files and make those changes for you. That's what Cowork and Claude Code are for.
What's in a Project
📚
The Knowledge Base — your brief pack
Upload files (PDFs, docs, CSVs, etc.) and they become permanently available to every
conversation inside the Project. Brand guide, research notes, past work, meeting transcripts —
anything Claude should know before you start talking. You can upload a substantial amount of
content, and when you exceed the limit, Claude automatically starts pulling in only what's
most relevant to each message — so you don't need to manage it manually.
📋
Custom Instructions — standing rules
Write instructions that shape how Claude behaves across every conversation in the Project.
This is where you'd set tone, role, output format, and conventions — anything you'd otherwise
have to say at the start of every chat. Example: "Always follow our design system naming
conventions. When reviewing copy, flag anything that contradicts the brand voice guide
uploaded here."
One thing to know about conversations within a Project: they don't see each other.
If you finish a great draft in one conversation, open a new one, and ask "can you continue where
we left off?" — Claude won't know what you mean. Each conversation draws from the same knowledge
base and instructions, but they're otherwise separate threads.
The workaround: if you want output from one conversation to carry into future ones, you can save
it to the Project's knowledge base manually.
Sharing a Project with your team
Projects can be shared with anyone else on your team account. When you share one, you choose
the access level: Can use means they can chat within the Project but can't
change its setup; Can edit gives full access to modify the instructions,
knowledge base, and settings.
A shared Project means the whole team is working from the same briefed-up version of Claude —
consistent behavior, consistent context, without everyone having to maintain their own copy.
Projects are private by default; you explicitly choose who gets in.
What designers use it for
🗃️
Client or brand workspaces
Upload brand guidelines, tone of voice docs, past work, and briefs. Every conversation for that client already knows the context — you just ask your question.
📐
Design system reference
Keep your component documentation, naming conventions, and usage guidelines in a Project so Claude can check anything against them, or help you write new entries that match the existing style.
🔬
Research synthesis
Drop in interview transcripts, survey results, and research notes. Ask questions, pull quotes, find patterns — without re-uploading every time.
🤝
Team shared context
A shared Project means everyone works from the same briefed-up Claude — consistent output, less repeated setup across the team.
Chat vs. Projects — when to use which
| Situation |
Use Chat |
Use a Project |
| One-off question or task |
✓ |
— |
| Recurring work with consistent context |
— |
✓ |
| Working from specific reference documents |
— |
✓ |
| Sharing context with teammates |
— |
✓ |
| Sensitive or confidential brainstorm |
✓ (use incognito) |
— |
| Ongoing client or project work |
— |
✓ |
Claude as an executive assistant — not just answering questions, but getting things done.
The mental model
Remember the limitation in Projects — that Claude hands you output like a file over Slack, and
you're the one who has to download it, upload it to Box, paste it into the doc, integrate it
wherever it needs to go? Cowork is where that constraint disappears.
With Cowork, that same tenured, well-briefed coworker now has direct access to your files and
your computer. You describe what you want done — "go through these five research files, pull
out anything relevant to our accessibility audit, and put together a structured report" — and
they go do it. They work across your files and folders, coordinate the steps themselves, and
come back when they're finished or when they've hit a decision only you can make.
Think of it less like a conversation and more like assigning a task. You give the outcome, not
the instructions.
Where to find it: Cowork lives in the Claude Desktop app (Mac and Windows) — not
in the browser. Look for the "Cowork" tab once you've got the desktop app open. You'll also need
to point Claude at the folders it's allowed to work with before it can touch your files.
What Cowork can actually do
📁
Read and write your local files directly
No uploading, no downloading. Claude works with files in the folders you've given it access
to — opening, editing, creating, renaming, and organizing them on your behalf.
🔀
Break complex tasks into parallel workstreams
For bigger jobs, Cowork can spin up multiple sub-agents working simultaneously on different
parts of the task, then pull the results together. You don't have to orchestrate that — it does.
🖥️
Use apps on your computer
Cowork can operate software on your desktop — navigating dashboards, working inside
spreadsheets, interacting with tools that don't have an API. It's using your computer the
way you would, just without you having to sit there.
🗓️
Run tasks on a schedule
Set up recurring tasks — a weekly summary, a daily file sort, a regular report — and Cowork
runs them automatically on your schedule, while your computer is awake and the app is open.
📱
Assign tasks from your phone
You can kick off a Cowork task from the Claude mobile app while you're away from your desk,
and your desktop handles the work. Come back to a finished deliverable.
What designers use it for
🗂️
File and asset organization
Point Claude at a messy downloads folder or a sprawling project directory and ask it to
rename, sort, deduplicate, or surface what's relevant. The kind of cleanup that always
gets deprioritized.
📋
Research synthesis and report drafting
Feed it a folder of source files — briefs, transcripts, PDFs — and ask for a structured
draft. Claude handles the assembly and synthesis; you handle the refinement.
🔁
Repetitive, multi-step tasks
Anything that's time-consuming but not technically complex — updating a set of documents
with new information, pulling consistent data from a batch of files, reformatting a
folder of deliverables to match a new template.
A few things to know before you start
👀
You stay in the loop
Cowork shows you what it's doing at each step. You can jump in to redirect it mid-task,
and it will always ask your explicit permission before permanently deleting any files.
🔒
You control what it can access
Claude only works with the folders you've explicitly shared with it. A good habit: create
a dedicated working folder for Cowork rather than giving it broad access to everything.
💻
Your computer needs to stay on
Cowork runs on your machine. If the desktop app is closed or your computer goes to sleep,
Claude can't work. Tasks resume when the app is back open.
Claude as a specialist contractor — building real, working things from your ideas.
The mental model
If Cowork is a generalist executive assistant, Claude Code is a specialist contractor —
someone brought in specifically to build things. Not to help you think through ideas or
hand you a document to integrate yourself, but to sit down and actually make something that
runs in a browser.
For designers at an agency working across many different client projects, jumping into
production codebases probably isn't the right entry point — and that's fine. That's not
what Claude Code is most useful for here. What it is useful for is everything
you'd want to build or explore on your own: interactive prototypes, animation experiments,
responsive behavior tests, side projects, small tools. Things where a Figma file or a
static mockup just isn't enough to answer the question.
You describe what you want. Claude Code creates the files, writes the code, and shows
you something real in a browser. You react to it, ask for changes, and iterate — the same
way you'd work with a developer, except the loop is just you.
Getting started — no terminal required
The biggest barrier for most designers is the assumption that Claude Code requires knowing
your way around a terminal. It doesn't have to. The Claude Desktop app includes Claude Code
in a visual interface — same capability, no command line. Download the desktop app, open
the Code tab, point it at a folder, and start describing what you want to build.
A good first project: create a new empty folder on your desktop, open it in Claude Code,
and say "build me a simple webpage with a card that animates in on scroll." You'll have
something to look at in a browser within minutes.
How to get it: Claude Code is included in your team account. Download the
Claude Desktop app at claude.ai/download, open it, and look for the Code tab.
If you'd rather start in a browser first, there's also a web version at claude.ai/code
— no installation needed.
What designers use it for
🎬
Prototyping animations and interactions
Scroll-triggered reveals, hover states, staggered entrance animations, custom cursor
behavior — these are nearly impossible to fully communicate in a static mockup. Claude
Code lets you build the real thing in a browser and iterate on it in plain English.
"Make the cards drift up as they enter" is a valid prompt.
📱
Testing responsive behavior in real code
Figma's responsive previews are approximations. If you want to know what a layout
actually does at 375px — whether the nav collapses right, whether the type scale holds
up, whether the spacing still works — Claude Code lets you see it in a real browser,
on a real device, and tweak it until it's right.
🛠️
Building side projects and personal tools
A portfolio site, a small tool you've always wanted, something you're building to
learn — Claude Code makes these tractable even without a development background.
You bring the vision and the taste; Claude handles the code.
✅
Validating specs before handoff
Designers sometimes spec things that are awkward to implement — unusual layouts,
specific timing curves, layered effects. Building a quick prototype in Claude Code
lets you pressure-test an idea yourself, so you arrive at the dev handoff with
more confidence, or knowing you need to rethink something.
📚
Learning front-end through doing
This has been one of the most valuable uses for designers who are curious about code
but don't have a background in it. You describe what you want, Claude Code writes it,
and you can ask it to explain what it did and why. You pick up how things work
naturally, at your own pace, with something real to point at — not a tutorial exercise.
A few things to know before you start
✅
It asks before making changes
By default, Claude Code asks for your approval before modifying files or running
commands. You're always in control of what actually gets changed.
🧠
It learns your project over time
Claude Code builds up memory of the project you're working in — your preferences,
decisions you've made, how you like things structured — and carries that into future
sessions automatically. The longer you work in a project, the less re-explaining you do.
📁
It works best with a folder to anchor it
Claude Code is most useful when it has a project folder to work in — even if that
folder starts completely empty. Think of the folder as the job site. Once it's there,
Claude can create files, build structure, and iterate within it.
Cowork vs. Claude Code — which one do I reach for?
| Situation |
Use Cowork |
Use Claude Code |
| Organizing or processing files |
✓ |
— |
| Synthesizing research into a report |
✓ |
— |
| Scheduled or recurring tasks |
✓ |
— |
| Building an interactive prototype |
— |
✓ |
| Prototyping animations or interactions |
— |
✓ |
| Testing responsive behavior in a browser |
— |
✓ |
| Building a side project or personal site |
— |
✓ |
| Learning front-end through experimentation |
— |
✓ |