Adoption Guide · Version 1.0 · June 2026
Getting started with Claude.
An adoption guide to the products, models, plans, and first steps — so your team can put Claude to work with confidence.
Section one
Introduction: what is Claude?
Claude is a family of AI assistants built by Anthropic.
You work with Claude through a few different products, each suited to a different kind of task: a chat assistant for everyday questions, an agent that completes multi-step work on its own, and a coding tool that reads and edits a whole software project.
SYRV put this guide together to help your team adopt Claude with confidence. It covers the three products (Claude, Claude Cowork, and Claude Code), the models behind them (Opus, Sonnet, Haiku, and the Fable flagship), the plans and pricing, the permission and admin controls, and a short walkthrough to get you started.
There are three ways to use Claude, arranged by how much autonomy you hand over — a conversation, a delegated task, or a coding agent.
1 · Claude
conversation
The chat assistant. You stay in the loop on every step — ask, review, refine.
2 · Claude Cowork
delegated task
You describe an outcome; Cowork does the office work and returns a finished result.
3 · Claude Code
coding agent
An agent that works across a whole codebase — reading, editing, running, committing.
In one breath: all three run on the same Claude models — what sets them apart is how much you hand off.
Section two
The Claude product family
2.1 · Claude — the chat assistant
Claude.ai is the conversational assistant most people start with. It runs in a web browser, as a desktop app for macOS and Windows, and on iOS and Android. You ask a question or give an instruction and Claude replies. It can search the web, read files you upload, analyze images, write and run code, remember context across a Project, and connect to tools such as Google Drive, Slack, and Microsoft 365 through connectors.
It suits drafting and editing, research, analysis, brainstorming, and quick questions where you want to stay in the loop on every step.
2.2 · Claude Cowork
Claude Cowork applies that same autonomy to everyday knowledge work that has nothing to do with code. Rather than a back-and-forth chat, you describe the outcome you want and Cowork does the work: organizing files, building and formatting spreadsheets, pulling data out of documents or screenshots, preparing reports, and handling recurring tasks on a schedule. It runs in the Claude desktop app, works with your local files and apps, can navigate a browser, and checks with you before any significant action.
in short
vs. Chat — Chat guides you step by step; Cowork completes the work autonomously and returns the finished result.
vs. Code — Claude Code targets software in the terminal/IDE; Cowork applies the same autonomy to office and knowledge work.
Availability — included on all paid plans (Pro, Max, Team, Enterprise); not on the Free plan.
2.3 · Claude Code
Claude Code is Anthropic's coding agent. It works across a whole codebase: reading and editing files, running commands, managing git, opening pull requests, and connecting to outside tools through the Model Context Protocol (MCP). It is the most hands-on of the three products and runs on the higher-capability models. Section 4 covers the ways to run it.
| Product | What it is | Best for |
|---|---|---|
| Claude | Conversational assistant (web, desktop, mobile). | Drafting, research, analysis, quick questions — staying in the loop. |
| Claude Cowork | Autonomous agent for office & knowledge work. | Handing off a multi-step task and getting a finished deliverable. |
| Claude Code | Coding agent across a whole codebase. | Building, editing, and maintaining software. |
Section three
The Claude models: Opus, Sonnet & Haiku
Claude comes in tiers from the same generation of training, each balancing speed, cost, and intelligence differently. As of June 2026 the widely available models are Claude Opus 4.8, Claude Sonnet 4.6, and Claude Haiku 4.5, with a premium flagship, Claude Fable 5, above the Opus tier for the most demanding work.
| Tier | Best for | Speed | Cost | Intelligence |
|---|---|---|---|---|
| Opus 4.8 | The hardest work — complex reasoning, long multi-step agentic tasks, deep coding & research. | Slower | Highest | Highest |
| Sonnet 4.6 | The everyday workhorse — drafting, summarizing, analysis, most coding. | Fast | Moderate | Strong |
| Haiku 4.5 | High-volume, simple, speed-sensitive tasks — classification, quick lookups, bulk work. | Fastest | Lowest | Good |
3.1 · Choosing a tier
As a rule of thumb: start on Sonnet for day-to-day work, move up to Opus when a problem is hard enough that quality matters more than speed, and drop to Haiku for simple work you run at high volume.
3.2 · What the models can do
Context window
how much it holds at once
How much Claude can hold in mind at once. Opus 4.8, Sonnet 4.6, and Fable 5 each have a 1-million-token window; Haiku 4.5 has 200,000 tokens.
Adaptive thinking
reasons as hard as needed
Claude decides on its own how hard to reason — thinking longer on complex problems and answering directly on simple ones.
Vision
reads images
All current models can interpret images: screenshots, charts, photos, and scanned documents.
3.3 · The flagship — Claude Fable 5
Claude Fable 5
Anthropic's most capable widely released model, sitting above the Opus tier for the most demanding reasoning and long-horizon agentic work, with always-on thinking and a 1-million-token context window.
Section four
Claude Code: the TUI vs. the app
Claude Code is the same engine wherever you run it; what changes is the interface and the workflow around it. The question we hear most is how the terminal version (often called the TUI) differs from the desktop and web apps, so that is where we will start.
4.1 · The terminal CLI (the "TUI")
This is Claude Code run in your terminal with the claude command — a text-based interface (TUI) right where developers already work. It offers the full feature set: every slash command, inline diffs, session history, scripting and piping, and it runs anywhere a terminal does — locally, over SSH, in a container, or in CI/CD. It is the recommended starting point and the most flexible surface.
4.2 · The desktop app & the web app
The Claude Code desktop app (macOS and Windows) and the web app at claude.ai/code wrap the same engine in a graphical experience and add capabilities that the bare terminal does not have:
Parallel sessions
sidebar
Run several tasks at once, each isolated on its own git branch.
Visual diffs
click to approve
Click-to-approve diffs and an integrated editor and terminal.
Background & cloud
runs on Anthropic's servers
Long-running tasks continue even with your laptop closed; the web app needs no local setup at all.
Mobile & scheduled
monitor anywhere
Mobile monitoring and scheduled/recurring tasks.
In one breath: the terminal runs anywhere and does everything; the apps add a GUI, parallel sessions, and cloud or background execution — most developers use both.
4.3 · IDE integrations
Claude Code also runs inside VS Code and the JetBrains IDEs (IntelliJ, PyCharm, WebStorm, and others) as an extension — a graphical panel with inline diffs, @-mentions of files, and a permission-mode selector, sharing your current editor selection and lint/type errors with Claude. The IDE extensions use the same CLI under the hood (install the standalone CLI separately to use the claude command in the IDE's terminal).
| Surface | Interface | Stand-out strengths | Best when |
|---|---|---|---|
| Terminal CLI | Text (TUI) | Full features; scripting; SSH/CI; runs everywhere. | Getting started; automation. |
| Desktop app | GUI (Mac/Win) | Parallel sessions; background & cloud tasks; visual diffs. | Many projects at once; long tasks. |
| Web (claude.ai/code) | Browser | Zero setup; cloud sessions; mobile monitoring. | No local setup; remote repos. |
| IDE extension | VS Code / JetBrains | Inline diffs; selection & error sharing. | You live in your editor. |
Section five
Permissions & controls
Permissions cover two separate things. One is how much Claude Code can do on your machine without asking, set by its permission modes. The other is who on your team can administer your organization, set by admin roles. The sections below take each in turn.
5.1 · Claude Code permission modes
Permission modes control whether Claude asks for approval before editing files or running commands. In the terminal, press Shift+Tab to cycle modes; in the apps and IDE extensions there is a mode selector.
| Mode | Auto-approves | Use for |
|---|---|---|
| default | File reads only — asks before any edit or command. | Getting started; sensitive work (safest) |
| acceptEdits | File edits and safe filesystem commands. | Iterating on code; review diffs after. |
| plan | Reads only — explores and writes a plan, stops before changes. | Exploring before committing to edits. |
| auto | Most actions, with an AI classifier blocking suspicious ones. | Longer tasks with fewer prompts (research preview). |
| dontAsk | Only pre-approved tools; everything else denied. | Locked-down, non-interactive CI/CD. |
| bypassPermissions | Everything, with no checks. | Isolated containers/VMs only — risky elsewhere |
safe
.git/, shell dotfiles, and tool config files) are always protected regardless of mode. Allowlists in settings.json let you pre-approve specific commands so you are not prompted repeatedly.5.2 · Team & Enterprise admin roles
On Team and Enterprise plans, administrators manage the organization through built-in roles:
| Role | What it grants |
|---|---|
| Primary Owner | One per organization; full control, can manage all data and members. |
| Owner | Full access; with the Primary Owner, the only role that can change member roles. |
| Admin | Administrative access, which can be scoped/delegated on Enterprise. |
| Member (User) | Standard participant. |
Role-based permissions let admins control which features and connectors specific groups can use, and delegate specific admin areas — Identity & Access, Billing, Analytics, Privacy, User Management, and Libraries — rather than granting blanket access. Enterprise additionally supports custom roles defined at the group level.
Section six
Plans & pricing
Claude subscriptions range from a free individual tier to custom Enterprise agreements. The tables below summarize the consumer and business plans (all prices USD; this is the Claude.ai subscription pricing, separate from the per-token Claude API).
6.1 · Individual plans
| Plan | Price / month | Usage | Highlights |
|---|---|---|---|
| Free | $0 | Limited baseline | Chat, web search, file uploads, MCP connectors, extended thinking. No Claude Code or Cowork. |
| Pro | $20 ($17 annual) | Standard paid usage | Everything in Free + Claude Code, Cowork, unlimited Projects, multiple models, Research. |
| Max 5× | $100 | 5× Pro usage | Higher limits & output; priority access to newest models. |
| Max 20× | $200 | 20× Pro usage | Highest individual usage; priority during peak demand. |
6.2 · Business plans
| Plan | Price / seat / mo | Minimum | Adds over the tier below |
|---|---|---|---|
| Team Standard | $25 ($20 annual) | 5 seats | Admin & billing tools, SSO + domain capture, role-based permissions, spend caps, enterprise connectors. Includes Claude Code & Cowork. |
| Team Premium | $125 ($100 annual) | 5 seats | Everything in Standard with substantially higher per-seat usage (5× a Standard seat). |
| Enterprise | Custom — contact sales | 20 (self-serve) / 50 (sales) | SSO/SAML, SCIM provisioning, audit logs, custom data-retention controls, HIPAA-ready (BAA), expanded context, analytics & compliance APIs, custom roles. |
Verify before you commit
Enterprise pricing is custom; a per-seat figure sometimes shown publicly is best confirmed directly with Anthropic sales, and Enterprise usage is typically billed at API rates rather than a fixed allowance.
Inclusion of Claude Code on Team Standard is per Anthropic's Help Center; confirm with your account team, as some third-party summaries state otherwise.
Max plan prices are web-subscription rates; mobile app-store pricing may differ.
Section seven
Getting started
7.1 · With Claude (chat) or Cowork
Create an account
Go to claude.ai (or install the desktop/mobile app) and create an account.
Choose a plan
Free to try it, or Pro to unlock Claude Code and Cowork.
Start working
Start a conversation, or open Cowork in the desktop app and describe a task to hand off.
Connect your tools
Connect tools (Google Drive, Slack, Microsoft 365) so Claude can work with your files.
7.2 · With Claude Code — your first 15 minutes
curl -fsSL https://claude.ai/install.sh | bash — then run claude to log in through your browser (Claude subscription or API key). On Windows, use the PowerShell installer; Homebrew and WinGet packages are also available.run
/init to create a CLAUDE.md so Claude follows your project's conventions every session.Handy first slash commands
| Command | What it does |
|---|---|
| /help | List all available commands. |
| /init | Create a CLAUDE.md describing your project for Claude. |
| /mcp | Add and manage external tool integrations (MCP). |
| /review | Review code changes. |
| /usage | Check your token usage. |
| Shift+Tab | Cycle permission modes. |
Section eight
Quick decision guide
8.1 · Which product?
| If you want to… | Use |
|---|---|
| Ask questions, draft, research, analyze — staying in the loop. | Claude (chat) |
| Hand off a multi-step office task and get a finished deliverable. | Claude Cowork |
| Build, edit, or maintain software across a codebase. | Claude Code |
8.2 · Which model?
| If the task is… | Choose |
|---|---|
| Simple and high-volume (classify, quick lookup). | Haiku 4.5 |
| Everyday work needing a balance of speed and quality. | Sonnet 4.6 |
| Hard — deep reasoning or long agentic work. | Opus 4.8 |
| At the absolute frontier of difficulty (premium). | Fable 5 |
8.3 · A recommended adoption path
Start on Pro
Start individuals on Pro to unlock chat, Cowork, and Claude Code in one plan.
Move to Team
Move teams of five or more to a Team plan for central billing, SSO, and shared admin controls.
Graduate to Enterprise
Graduate to Enterprise when you need SAML/SCIM, audit logs, data-retention controls, or compliance commitments such as a BAA.
Tune permissions as trust grows
Keep Claude Code in default or plan mode while learning; adopt acceptEdits and allowlists as trust grows.
Section nine
Key terms & resources
9.1 · A few terms you'll hear
| Term | In short | Plain-language meaning |
|---|---|---|
| Model | the AI itself | The underlying AI (e.g. Opus, Sonnet, Haiku) that generates responses. |
| Agent | works on its own | Claude working autonomously across multiple steps to finish a task — the idea behind Cowork and Claude Code. |
| Prompt | your instruction | What you ask or instruct Claude to do. |
| Token | unit of text | The unit Claude reads and writes in (roughly ¾ of a word); usage and pricing are measured in tokens. |
| Context window | how much it holds | How much text Claude can consider at once — e.g. 1 million tokens on Opus and Sonnet. |
| MCP | tool connector | Model Context Protocol — the standard that lets Claude connect to outside tools and data (Drive, Slack, GitHub, and more). |
9.2 · Official documentation & product pages
| Resource | Where |
|---|---|
| Claude plans & pricing | claude.com/pricing |
| Claude Cowork | anthropic.com/product/claude-cowork |
| Claude Code documentation | code.claude.com/docs |
| Claude Code permission modes | code.claude.com/docs/en/permission-modes |
| Models overview | platform.claude.com/docs/en/about-claude/models/overview |
| Roles & permissions | Claude Help Center (support.claude.com) |
June 2026