← Hub / 07 Cheat Sheet
07 Cheat Sheet

The Four Claudes Cheat Sheet

6 min read By Scott

There are four Claudes now, and most people are still only using one.

Chat thinks. Cowork does. Code builds. Design creates. Picking the wrong one is why the output keeps coming back generic. Here is when to reach for each, and how to chain them so the work actually leaves the chat window.

Most people are stuck on Chat

For a long time Chat was the only Claude, so that is the one almost everyone defaults to. It is also the wrong tool for roughly half of what people are pushing through it. Running EDGE and becomingscott in parallel with ADHD means leverage is the difference between things shipping and things stalling, and all four of these now sit in my daily rotation.

There are four distinct Claude products now, built for completely different jobs. Chat is the thinking partner. Cowork is the agent that takes action on your computer. Code reads and writes real code in your repo. Design generates the visual layer. When the output keeps coming back flat, it is usually because the thinking tool is being asked to do the building.

The real unlock is not any single Claude. It is chaining all four, using each one for what it actually does and handing the output to the next.

What follows is the cheat sheet. A side by side of what each Claude does, what it can access, when to reach for it, an example prompt per tool, and the chained workflows that combine all four.

The four at a glance
Think
Chat
The thinking partner you already know. Writing, brainstorming, analysis, and working a hard decision out loud until it has shape.
Do
Cowork
An agent that takes action on your machine, opening files, driving your browser, processing data, and running on a schedule while you are elsewhere.
Build
Code
Builds real apps, scripts, and automations from plain English, working directly inside your terminal, VS Code, or JetBrains.
Create
Design
The visual layer for decks, mockups, layouts, and 3D scenes, generated from a description and then refined element by element.
Claude Chat, the thinking partner
What it does

The conversational Claude, built for writing, brainstorming, analysis, decisions, learning, and chewing through documents. You talk through an idea and it pressure-tests the weak parts. You paste in a doc and it summarises, then asks the questions you forgot to ask.

What it can access

Web search, uploaded files including PDFs, docs, images, and code, Projects for persistent custom context that travels with you across chats, and Artifacts for documents and code that live inside the conversation.

When to reach for it

Drafting copy, working through a hard decision, getting up to speed in a new domain quickly, pressure-testing a strategy, or synthesising a messy pile of notes. Anywhere you would otherwise spend an hour at a whiteboard or on the phone to a smart friend.

Example prompt
"I'm trying to decide between [X] and [Y]. Steelman both sides at the level of a senior operator. Then pressure-test my current lean toward X. What am I underweighting?"
Claude Cowork, the agent that does
What it does

Operates your computer for you. It clicks through the browser, opens and edits files, runs multi-step workflows end to end, processes data at volume, and runs the same routine on a schedule. Where Chat tells you what to do, Cowork is the one that actually does it.

What it can access

Your screen and browser, local files and folders, and scheduled task runners. If a person could do it by clicking around your desktop, Cowork can do it without you sitting there watching.

When to reach for it

Repetitive multi-step workflows, pulling data out of tools that have no usable API, cleaning and renaming a folder of files, or running the same checklist every Monday morning. Anywhere you would otherwise hire a VA or block out an afternoon to just get through it.

Example prompt
"Open my inbox, pull every email from [vendor] in the last 30 days, summarise them into one doc on my desktop, and put the action items in a checklist at the top."
Claude Code, apps and automations from plain English
What it does

Reads, writes, and refactors actual code in your repo. It builds features, scripts, automations, and entire apps from a plain English description, runs the tests, debugs the failures, and ships the result. It lives in your terminal or as an extension inside VS Code and JetBrains, sitting where the work actually happens.

What it can access

Your repo, your local environment, your packages and dependencies, your tests, and your deploy commands. It can grep, edit, run, and commit, which is the whole loop a developer runs, only now you are reviewing the diff instead of typing it.

When to reach for it

Shipping a feature, refactoring a tired module, debugging something gnarly, building the internal tool you would otherwise pay a SaaS for, automating a dev workflow, or spinning up a quick script that pulls one number from one place and posts it somewhere else. Anything code-shaped, basically.

Example prompt
"Build me a script that pulls my Stripe revenue for the current month, compares it to the previous month, and posts the diff to Slack every Monday at 8am."
Claude Design, visual output from a sentence
What it does

Generates visual assets directly from a description. Pitch decks, landing page mockups, social graphics, brand explorations, 3D scenes, and product visualisations all sit inside its range. You describe what you want and Design produces it, then sits there waiting for you to push back on the bits that are not quite right.

What it can access

Your prompt, any reference images you upload, the brand guidelines you hand it, and an iterative loop where you point at a specific element, ask for a change, and watch it revise without losing the rest.

When to reach for it

Pitch decks, landing page mockups before you commit to building the real page, social tiles, and product mockups for a Figma file you have no patience to draw. Any visual where the prompt is clearer in your head than the pixels are, and where you would otherwise outsource it or lose an hour pushing shapes around in Canva.

Example prompt
"Make a 6-slide pitch deck for a D2C supplement brand targeting day traders. Brand colours: deep forest green, cream, sage. Editorial magazine feel. Big typography. Section dividers between slides."
Chained workflows, all four together
Ship a landing page in one afternoon
ChatTalk through the offer and the audience until the headline is sharp, the body copy is drafted, and the CTA is locked.
DesignMock the visual layout off the locked copy and iterate on the hero, the typography, and the rhythm of the sections.
CodeBuild the real page in clean HTML and CSS, dropping the locked copy in as it goes and matching the mockup beat for beat.
CoworkPush it live, run a Lighthouse check on the production URL, and schedule the launch post in your scheduler of choice.
Run a research project end to end
ChatFrame the question, scope the research, and draft an outline of what you are actually looking for.
CoworkPull the actual data by scraping sources, downloading reports, and organising it all into one tidy folder.
ChatSynthesise what Cowork pulled, build the argument, and surface the contradictions you would otherwise miss.
DesignTurn the synthesis into a one pager or short deck you can actually share with a stakeholder without apologising for it.
Ship an internal automation tool for yourself
ChatTalk through what you actually need, what failure modes are worth designing around, and what done looks like.
CodeBuild the script or small app and wire it cleanly to the inputs and outputs you sketched in Chat.
CoworkSet up the recurring run and hook the output to your inbox, your Slack, or your calendar, wherever it actually needs to land.
ChatQuarterly, paste in the logs and the failures and decide what to tweak. The tool gets sharper every time you do this.
The five handoff patterns
01

Chat → Code

Use Chat to think through the spec first, covering what the tool does, what it deliberately does not, and where it is likely to fail. Then hand a clean written spec to Code and let it build against that. The output is dramatically better than asking Code to figure out both the design and the implementation in one go.

02

Cowork → Chat

Cowork is excellent at pulling raw inputs and ordinary at synthesis. Let it gather the data and dump it into one doc, then take that doc into Chat for the actual thinking. One agent pulls, the other processes, and the result is sharper than either could produce alone.

03

Design → Code

Mock the visual in Design first and only lock in the layout once it actually looks right. Then hand the mockup and the locked copy to Code and ask it to build the real thing in HTML and CSS. Skipping the Design step is the reason most landing pages built by devs end up looking engineered rather than designed.

04

Chat → Cowork

Chat decides what needs to happen, including the sequence, the safety checks, and the edge cases worth catching. Cowork then executes those steps in order without needing you in the chair. This is how a single twenty minute conversation turns into half a day of finished work done while you are somewhere else entirely.

05

The anti-pattern

The anti-pattern is trying to do all four jobs inside Chat. It can fake the other three convincingly enough to feel like progress, but the output stays trapped in the chat window with no file on disk, no app shipped, and no deck anyone can actually open. Chat is the thinking partner. Cowork, Code, and Design are the hands that turn the thinking into something real.

Want this mapped to your situation?

Book a call. Happy to do the work directly, advise you through it, or connect you with the right person in my network. Every problem leaves with a path forward.

Book a Call →