The Claude Stack.

Four products, one brain. Claude.ai for thinking. Cowork for finishing. Code for anything technical. Design for visuals. For women ready to put the whole stack to work.

By Anne Murphy

About This Preamble
(it's kind of a lot)
A resource for anyone who wants an on-ramp to the whole Claude stack. Share it with your friends. Bookmark it. Come back to it when you add the next product.
Read the companion guide, The Copilot Stack → Explore She Leads AI →
01

Why This Reference Exists

Four products, one brain. Each earns its place in the daily work.

Anthropic now ships Claude across four distinct surfaces, each one earning its place in how a working woman runs her week. They share a brain, so the patterns you learn on one apply straight across to the others. Use the whole stack and the daily work changes shape.

The shorthand for the four products is one sentence. Claude.ai is the conversation. Claude Cowork is the coworker who finishes the job on the files you have. Claude Code is the operator inside your terminal who handles anything a developer would handle. Claude Design is the studio that takes a sentence and gives you back a working visual.

Use the whole stack and the daily work changes shape.

Skills, Projects, and Memory stitch them together. Learn the patterns once and the same habits apply across every surface. That is the whole game.

This preamble pulls the canonical Anthropic documentation into the language we already speak. Use it as a reference. Use it as a Saturday morning read. Hand it to a friend who keeps asking why everyone is talking about Cowork.

02

The Big Picture

Four products. One model family. One brain.

The four products share a brain. The current Claude model family powers all of them, and choosing the right model counts more than most people realize.

Claude Opus 5 is the one to reach for on hard reasoning, complex coding, and work that runs long without you watching. It arrived in July 2026 as a thoughtful and proactive model at half the price of Fable 5, and it is a clear step up from Opus 4.8 on deep thinking and on agentic work that spans many steps.

Claude Sonnet 5 is the daily driver, and it now lands close to Opus quality on coding and agentic work. Use it for everyday building where speed and cost still count.

Claude Haiku 4.5 is fast and inexpensive. Best for bulk tasks, quick questions, and automations where speed counts more than depth.

Opus 5 and Sonnet 5 each have a one-million-token context window as standard, which changes what is possible with long documents and large codebases. Above all of them sits Claude Fable 5, the most capable model Anthropic has released, built for the most demanding reasoning and the longest autonomous runs.

ProductWhere It LivesBest ForWho Uses It Most
Claude.aiBrowser, iOS, Android, desktop appConversational work, research, writing, analysis, light codingEveryone
CoworkDesktop app on Mac or Windows, Tasks tabMulti-step jobs on the files you have, scheduled tasks, pluginsOperators, analysts, project leads
CodeTerminal, VS Code, JetBrains, web, mobileAnything technical — code, automation, agents, integrationsDevelopers and technical operators
Designclaude.ai/designPrototypes, decks, one-pagers, marketing visuals, UI mockupsMarketers, designers, founders
03

Claude.ai — The Conversation Layer

The browser product. Where most people start, and where the daily thinking gets done.

Claude.ai is the consumer chat product. It runs in a browser or the native desktop app, and it is where most people start. What used to be a text-in, text-out chat box is now closer to a workbench — Claude can read your files, search the web, create documents, run code, remember you across sessions, and connect to outside systems.

Use Projects for anything you will revisit. The setup pays back inside a week.

Projects

Group related conversations under a single set of custom instructions and uploaded files. Everything in a Project inherits its context. Free users get up to five. Pro and above get unlimited. Use a Project when the same context applies across many chats — a client, a program, a research thread.

Memory

The feature that makes Claude feel like it knows you. Available on Free and up, Memory stores facts and preferences across conversations — how you write, which clients you work with, what tools you use. Since July 2026 it keeps individual categorized entries updated as you talk rather than one rolling daily summary, which makes it far easier to find a wrong fact and delete it. View and edit in Settings > Memory. Scoped per workspace on paid plans.

Artifacts

The panel on the right of your chat where Claude renders code, documents, diagrams, and interactive HTML. They are live — Claude can update them in place, and you can copy or download them. The working draft of whatever you are making, side by side with the conversation.

Skills

Reusable capabilities. Anthropic ships built-in skills for Word docs, Excel, PowerPoint, and PDFs. You can add custom skills in Settings > Capabilities > Skills by uploading a folder with a SKILL.md file. They only load when relevant, so they do not crowd the context window. She Leads AI has sixty-plus custom skills already. Enterprise plans can now scan third-party skills and plugins for malicious content before anyone installs them.

Connectors

Live access to outside systems — Slack, Notion, Google Drive, Asana, Fireflies, Kit, Canva, GitHub. Claude can search, read, and write back when granted permission. The Microsoft 365 connector gained write access in July 2026, so Claude can now send mail and manage calendar entries rather than only reading them. Admins on Team and Enterprise plans control which connectors are available.

Research Mode

Multi-step web hunt with citations. Claude plans, fetches, cross-checks, and returns a report. Use it for deep dives that would normally take you an hour in browser tabs. Available on Pro and above.

Computer Use

Lets Claude control your computer — opening files, clicking, typing, running tools. Powerful but slower than native tools, and always asks permission before doing anything destructive. Available on Pro and Max.

Excel, PowerPoint, Mobile

Add-ins put Claude directly inside Microsoft 365. In Excel, Claude writes formulas, cleans data, and assembles models. In PowerPoint, it makes and edits slides. The iOS and Android apps handle chat, voice input, image analysis, and Cowork tasks fired from your phone that execute on your desktop.

Practices
  • Be specific about what done looks like. Audience, length, tone, format, examples. Vague prompts give vague answers.
  • Paste in your raw material. Claude is better with your raw transcript, email, or data than with a description of it. Let Claude summarize for you.
  • Use Projects for anything you will revisit. The setup pays back inside a week.
  • When Claude is wrong, point at the sentence and say which rule it broke. Specificity beats "make it better."
  • Switch models with the task. Opus for hard thinking. Sonnet for everyday work. Haiku when you need speed. Start a new chat for a new task.
  • Check Memory monthly. Stale memory causes weird answers. Open Settings > Memory and clean it out.
04

Claude Cowork — The Coworker

Claude takes a multi-step assignment, does it on the files you have, and delivers the finished thing.

Cowork turns the Claude.ai desktop app into a working partner. Instead of answering questions, Claude takes multi-step assignments, executes on the files you have, and delivers a finished thing — a spreadsheet, a folder of renamed documents, a report pulled from five sources.

Chat answers. Cowork finishes. In chat, Claude describes how to reorganize a folder. In Cowork, Claude reorganizes the folder. It has file system access (only to folders you connect), can break a job into subtasks, and can run on a schedule. This is the mode most She Leads AI agent skills are made for.

Chat answers. Cowork finishes.

Cowork went generally available on macOS and Windows in April 2026, then opened up on web and mobile in July with remote sessions. Sessions and files now save to your Claude account and follow you across devices, work keeps going after you close the laptop, and a scheduled task fires whether or not any machine of yours is awake.

To get started, install the Claude Desktop app on macOS or Windows, sign in with Pro or above, click the Cowork tab, and connect a folder. Claude can only touch folders you grant access to. Describe the outcome you want. Claude proposes a plan, asks for approval, and executes.

Tasks Mode

Where Cowork lives. Reads, edits, and creates files in folders you connect — no manual upload and download loop. Breaks work into coordinated subtasks. Long tasks do not blow the context window because Cowork spawns focused sub-agents.

Native Office Output

Properly formatted Excel with formulas, PowerPoint with brand colors, Word with proper styles. Cowork makes and edits these as themselves, not as copies pasted into a chat.

Scheduled Tasks

Daily briefings, weekly reports, Monday-morning scans. Set it once and forget it. The cleanest path from "I do this every week" to "this runs on Mondays without me."

Plugins

Pre-packaged bundles of Skills, Connectors, and Sub-agents for a specific job. Anthropic has open-sourced eleven — Sales, Finance, Legal, Marketing, HR, Engineering, Design, Operations, Data Analysis, Customer Support, Product Management. Install one and the right skills activate immediately. She Leads AI has internal plugins, too.

Project Workspaces

Persistent memory, files, and instructions just for that body of work. The Cowork equivalent of a Claude.ai Project, scoped to long-running operational threads.

Anywhere Sessions

Start on the phone, pick it up on the laptop, check it from a browser. Remote sessions run on Anthropic's side rather than your desktop, so a job you kicked off from the car is still running when you sit down. Rolling out by plan, beginning with Max.

Practices
  • Describe outcomes, not steps. "Organize this folder so I can find client files by quarter" beats "make a 2026-Q1 folder, then a 2026-Q2 folder."
  • Approve the plan before execution. Cowork shows its plan. Read it. A ten-second sanity check saves twenty minutes of wrong work.
  • Scope the folder tightly. Do not hand Claude your whole Documents directory. Give it the one folder relevant to the task.
  • Schedule anything you do weekly. If you find yourself running the same Cowork task on Mondays, put it on a schedule.
  • Install role-based plugins. Marketing, Sales, and Operations ship with skills most She Leads AI team members would otherwise have to assemble from scratch.
05

Claude Code — For Anything Technical

An agentic environment for code, automation, agents, and integrations. The home of the She Leads AI agent stack.

Claude Code is an agentic coding environment. It runs in your terminal, VS Code, JetBrains IDE, or on the web. It reads your files, runs commands, makes changes, and works through problems autonomously while you watch, redirect, or step away. Even if you do not call yourself a developer, Claude Code handles any task a developer would handle — automations, integrations, data pipelines, custom agents, She Leads AI skill development.

Claude Code has a context window — a limited amount of information it can keep in view at once. Every file it reads, every command it runs, every message fills that window. The single most valuable skill in using Claude Code well is managing that context. Tell it exactly what counts. Clear it between tasks. Use tools that keep exploration from eating the whole budget.

The single most valuable skill in using Claude Code well is managing context.

CLAUDE.md

A markdown file at the project root that Claude reads at the start of every session. Stores persistent project context Claude cannot guess — the test runner you use, the code style that differs from defaults, the repo etiquette, the architectural decisions. Treat it like code. Prune ruthlessly. Check it into git.

Skills

Directories in .claude/skills/ with a SKILL.md at the top. YAML frontmatter names the skill and describes when to use it. Claude loads them on demand, so they do not clutter every conversation. Invoke directly with a slash command or let Claude find one from the description.

Subagents

Specialized assistants in .claude/agents/ with their own tools and context. Use one when a task needs to read many files without cluttering your main conversation — a security review on a PR, an investigation that would otherwise pollute the context, a writer-and-reviewer pattern.

Hooks

Scripts that run automatically at lifecycle points — SessionStart, PostToolUse, PermissionRequest, Stop. Unlike CLAUDE.md instructions, which are advisory, hooks are guaranteed. Use them when something must happen — run the linter after every edit, block writes to migrations, require a commit message format.

MCP Servers

Anthropic's standard for connecting Claude to external systems. Run claude mcp add to wire up Notion, Slack, Figma, Fireflies, databases, Kit. Once connected, Claude can read and write in those systems without you writing custom integration code.

Plan Mode and Auto Mode

Plan Mode — Claude explores without making changes and returns a plan for you to approve. Auto Mode — a classifier reviews commands and blocks only risky ones, so you do not click "approve" fifty times. Accept Edits Mode pre-approves edits for a long implementation run.

Practices
  • Run /init on every new repo first. Generates a starter CLAUDE.md you can prune.
  • Use /clear between unrelated tasks. The command that compounds the fastest. A fresh context with a precise prompt beats a long context with accumulated corrections.
  • Parallel sessions on big work. Claude --worktree spins up three to five sessions in isolated git worktrees. Strong for large migrations or experimentation.
  • Let Claude interview you for big features. "Interview me with AskUserQuestion before you plan this." Claude asks the obvious and non-obvious questions, then writes SPEC.md.
  • Rewind freely. Every action checkpoints. Shift+Escape opens a menu to restore code, conversation, or both. Try risky things cheaply.
  • Scope investigations narrowly or use a subagent. "Investigate our auth system" with no scope causes Claude to read 200 files. Subagents keep the exploration out of your main context.
06

Claude Design — Visuals From a Conversation

Polished visuals from a chat panel and a live canvas. Launched April 17, 2026.

Claude Design launched April 17, 2026 as an Anthropic Labs product. It lets you make polished visual work — designs, prototypes, presentations, one-pagers, marketing assets, UI mockups — by talking to Claude. Available to Pro, Max, Team, and Enterprise subscribers at claude.ai/design.

For She Leads AI, this is the fastest path from idea to shareable visual we currently have. It shortens (sometimes replaces) the gap between "we need a one-pager" and "here is the one-pager." For CREATE 2026, Social Saturday, companion cards, and sponsorship decks — learn it well.

A chat panel paired with a live canvas. Describe what you need, and Claude makes the first version.

The interface is a chat panel paired with a live canvas. Describe what you need. Claude makes a first version on the canvas. Refine through chat-based edits, inline comments on specific elements, or adjustment controls for spacing, color, and layout.

Chat-Based Edits

"Make the hero bolder." "Swap the palette for something warmer." Broad, conversational changes Claude applies across the whole design.

Inline Comments

Click an element on the canvas, leave a comment. Claude applies the change just to that element. The precision pass after the conversational pass.

Design System Import

Link your codebase. Claude picks up your brand's components and uses them automatically. Reference the design system by name in chat — "use our primary button" — and Claude knows what you mean.

Web Capture

Point at a competitor's site. Claude grabs elements directly. The fastest way to mark up a "more like this" reference.

Organization-Scoped Sharing

Keep private, share view-only, or grant edit access so teammates can chat with Claude together on the same design. Live collaboration where the design and the conversation share a canvas.

Handoff to Claude Code

When the design is ready, send it to Claude Code as a single instruction to ship the production version. The seam between design and engineering disappears.

Practices
  • Start simple, layer complexity. Get the structure right before asking for polish. Layout first, then content, then styling.
  • Be specific about feedback. "Make the CTA more prominent" beats "make it pop." Reference elements by name.
  • Ask for variations. "Give me three alternate versions of the header" surfaces options you would not think to request.
  • Reference the design system by name. If you have linked a codebase, say "use our primary button" instead of describing it.
  • Think responsive early. Mention desktop, tablet, and mobile breakpoints up front so Claude plans for them.
07

How They Plug Together

Skills, Projects, Memory, MCP, Subagents, Hooks — match the mechanism to the question.

Claude has several ways to keep persistent context or extend its abilities. They overlap. The shortcut is to match the mechanism to the question.

You type the same instructions repeatedly.
Make a Skill

Reusable procedures. Claude loads them on demand. Available across Claude.ai, Cowork, and Code.

Claude needs to know something about YOU.
Use Memory

Personal preferences, ongoing context. Persists across every Claude.ai conversation. Scoped per workspace on paid plans.

Claude needs to know something about a WORKSTREAM.
Use a Project

Client-specific or program-specific context. Every chat in that Project inherits the same instructions and files.

Claude needs to reach an outside system.
Add a Connector or MCP

Connectors in Claude.ai and Cowork. MCP servers in Claude Code. Both give Claude live access to Slack, Notion, GitHub, Drive, and dozens more.

A task deserves isolated focus or different tools.
Spin up a Subagent

Specialized agents with their own context window. Strong for security reviews and investigations that would otherwise crowd your main thread.

Something must happen every time without fail.
Write a Hook

Claude Code only. Scripts that run automatically at lifecycle points. Guaranteed, not advisory.

08

Ten Rules for Working With Claude Well

These compound. Internalize them and the quality of everything Claude makes for you climbs.
01

Give Claude a Way to Verify

Tests, screenshots, expected outputs, a bash command that proves success. The single habit that makes the biggest difference.

02

Be Specific About Done

Audience, length, tone, format, examples. Vague in, vague out. Tell Claude what the finished thing looks like before asking for it.

03

Explore, Plan, Then Make

For anything non-trivial, use Plan Mode in Claude Code or ask Claude to draft a plan before executing. Approve, then go.

04

Clear Context Between Tasks

Long sessions with accumulated noise perform worse than fresh sessions with a precise prompt. The /clear command pays back every time you use it.

05

Reference Files, Do Not Describe Them

Paste, drag, attach, or use @path. Claude is better with the raw material than with your summary of it.

06

Point at Examples

"Match the tone of this existing post." "Follow the structure of this existing skill." Patterns beat descriptions.

07

Correct Early

When Claude goes off track, redirect in the next turn. Do not wait three turns and then untangle a mess.

08

After Two Corrections, Start Fresh

Slash-clear, rewrite the prompt, incorporate what you learned. Do not pile corrections on top of each other.

09

Pick the Right Product

Claude.ai for thinking. Cowork for multi-step work. Code for anything technical. Design for visuals. Match the tool to the job.

10

Treat Context as a Budget

Every file read costs context. Every command costs context. When you are running hot, use subagents, /compact, or /clear. Performance degrades as context fills.

09

Where to Start

Pick your row. The four products mapped to the roles women are already running.

Find your row. Start there.

For Operators
  • Cowork plus the Marketing, Operations, and Sales plugins on every Pro-or-above account.
  • The named She Leads AI agents — Sheila, Romy, Rena, Marie, Margaux, Scout, Pepper — run best in Cowork Tasks with Notion, Fireflies, and Slack connectors live.
  • Scheduled tasks for the weekly rhythms — Monday morning pulse, Friday SEO audit, daily Social Saturday status check.
For Content and Brand
  • Claude.ai Projects for long-running content — one per major workstream (CREATE 2026, Social Saturday, The SLAI Effect, the podcast).
  • Claude Design takes a meaningful share of the Canva work for one-pagers, companion cards, and pitch visuals. Learn it well before CREATE.
  • She Leads AI brand enforcement fires on every outward-facing draft and is wired into most content skills already.
For Anything Technical
  • Claude Code is where skills get made, audited, or rebuilt. The creating-skills and skill-creator tooling lives there.
  • Run /init on any repo first. Prune CLAUDE.md down to the rules that count.
  • Subagents for security and audit passes. The /security-review and /review slash commands are built in.
For the Community
  • Claude.ai Free is the on-ramp. Everyone can start there and feel the difference.
  • Memory and Projects are the two features that pay back in any Social Saturday. They are the difference between Claude as a chatbot and Claude as a coworker.
  • Claude Design is the flashy demo — produces a visible win immediately. Strong on any live demo.
10

Sources and Further Reading

Pulled from Anthropic's official documentation and product blog. Re-crawl quarterly to stay current.

Claude.ai

  • Claude.ai features — the consumer chat product
  • Plans and pricing — Free, Pro, Max, Team, Enterprise
  • Release notes — what is new and when
  • Projects — bundled context per workstream
  • Skills in Claude — what they are and how they fire

Cowork and Code

  • Get started with Claude Cowork — install, sign in, first task
  • Cowork launch blog — Cowork brings Claude Code power to knowledge work
  • Plugins in Cowork — install, customize, build new ones
  • Claude Code quickstart — CLI, IDE, first run
  • Best practices for Claude Code — canonical patterns
  • Claude Code power user tips — parallel sessions, rewind, interview mode

Design and Concepts

  • Introducing Claude Design — Anthropic Labs launch post
  • Get started with Claude Design — the chat-plus-canvas interface
  • Design plugin for Claude — how Design plugs into Cowork
  • Skills explained — Skills vs. Prompts, Projects, MCP, Subagents
  • Extend Claude with Skills — the canonical Skills doc
The She Leads AI Society is the membership for women in AI. It includes the Build, Launch, Scale Lab, weekly Social Saturdays at no cost for all women, Member Jam Sessions, the She Leads Reads Book Club, and a 24/7 community on Mighty Networks. $787 a year (save 15%) or $77 a month, everything included.
Join the Society →
CREATE 2026 — the human-centered AI conference for women. October 16–18, Salt Lake City.
Gather With Us →
Social Saturday is made possible by the She Leads AI Society, our membership program for human-centered AI practitioners.
Anne Murphy
Anne Murphy

Anne Murphy is founder of She Leads AI, an AI Academy grounded in powerful community, where she has educated more than 4,000 professionals in responsible AI use. She is co-author of the Framework for Responsible AI in Fundraising and co-host of The Daily AI Show. Anne is also co-founder of Moxxee, an ethical, relational CRM built by fundraisers for fundraisers, and CEO of Empowered Fundraiser Consulting, with 35 years of experience and more than $10 billion in fundraising campaigns across nonprofit and higher education.

GenX mom, backpacker, Midwest roots, Pacific Northwest vibes.

Built by Anne with her matriarchal agentic leadership team.

Anne on LinkedIn Anne on TikTok She Leads AI She Leads AI on LinkedIn

Keep Going With She Leads AI

She Leads AI is the AI education and community platform for women in human-centered AI — an AI Academy grounded in powerful community.

Explore She Leads AI → Read The Copilot Stack →