Practical Onboarding Guide
Getting Started with Claude & Cowork
A practical guide to getting Claude and Cowork working for you.
Most people use AI as a smarter search box. That is the smallest version of it. Claude works far better treated as a capable teammate: it remembers your context, works inside your tools, and finishes multi-step jobs while you are busy with something else. This guide gets you set up first, then shows you how to move from quick questions to real work that runs on its own.
How to use this guide
1 Install Claude
- Log in and download the desktop app: claude.com/download. Desktop is still the fullest experience. It is the only surface with direct local file access, browser control, and computer use. Cowork is rolling out on web (claude.ai) and mobile too, so you can start or check on work from anywhere. Use desktop when a task needs your files, browser, or apps.
- Install the mobile app too and set up Dispatch so you can hand off tasks from your phone that need your desktop: local files, browser, installed apps. Your computer must stay on and the desktop app open while it works. For tasks that do not need your desktop, Cowork can also run as a cloud session that keeps going even if your computer is off.
- Claude Pro is $20/month per seat and remains the best value in AI today. Max plans ($100/month for 5x usage, $200/month for 20x) raise your usage limits and give you priority access to new models and features if you start hitting caps on Opus or Fable.
- Turn on Code execution and file creation in Settings > Capabilities. This one setting turns on file creation, skills, and most of what is in this guide. If a skill or file request does not seem to work, check here first, or ask your org owner, since Team and Enterprise can control this centrally.
- Install Claude in Chrome too, if your work touches websites. It is a separate browser extension with its own permissions. Claude can click, type, and navigate pages for you.
2 Complementary Subscriptions
Claude is the workhorse. These tools cover the gaps depending on your needs:
| Tool | Cost | Why I use it |
|---|---|---|
| Claude Pro | $20/mo. | Primary productivity tool. The best value in AI today. |
| Google Gemini | $20/mo. | NotebookLM, image generation (Nano Banana), and video capabilities. |
| Perplexity | $20/mo. | Detailed research that requires full fact-checking and source references. |
3 Give Claude Memory and Context
Claude now remembers context across conversations automatically, for every plan including Free. The biggest lever for better output is keeping that memory, and any context you add to it, accurate and current.
- Review, edit, or clear what Claude remembers anytime in Settings > Memory.
- Import memory from ChatGPT, Gemini, or Copilot directly: claude.com/import-memory. It carries over your preferences and context (not your full chat history) and takes about 24 hours to fully absorb.
- Customize your profile inside Claude (name, role, how you like responses formatted) so it carries across chats.
- For ongoing work, keep a Markdown (.md) memory file that captures your role, your businesses, your preferences, and key facts. It is portable across tools, easy to hand to a teammate, and gives Cowork projects (which keep their own separate memory) a clean starting point. Keep one per business or project and load the relevant one when you start.
# About Me - Role: President & Co-founder, Auditrol (agentic AI for bank/CU risk) - I also run CBCS Capital (family office), Carrollwood, and Lucid Bots # How I like responses - Concise, bullet format. Side-by-side tables for comparisons. - Always include source links. I will check your work. # Current priorities - Auditrol: new-customer acquisition and industry awareness
Tip. Ask Claude to help you build this file. Paste in a few emails or a bio and say, "Turn this into a memory file I can reuse."
4 Choose the Right Model
Match the model to the job to balance capability, speed, and cost. You can switch models mid-project.
Figure 1 — Depth versus speed. Move up only when the problem demands it.
| Model | Best for | Notes |
|---|---|---|
| Fable 5 | The hardest, highest-stakes problems: novel reasoning, research-grade work | Anthropic's top tier. Most capable and most expensive; reserve for work where being wrong is costly. |
| Opus 5 | Deep thinking, planning, complex multi-step problems, agentic work | Strong reasoning at a more accessible price than Fable. Use for genuinely hard planning or analysis. |
| Sonnet 5Default | Day-to-day execution, drafting, analysis, most Cowork tasks | The recommended default. Strong balance of speed, cost, and quality. |
| Haiku 4.5 | Fast, lightweight, high-volume tasks; quick lookups | Fastest and lowest cost. Less capable than Opus or Fable on complex work. |
5 Connect Your Tools
Connecting Claude to the systems you already use is what turns it from a chatbot into a teammate. A connector is a secure, permissioned link between Claude and an outside tool. It is what lets Claude read your Gmail or write to your CRM instead of just talking about it. Skills, connectors, and plugins now live in one directory inside Claude. A plugin bundles connectors, skills, and sub-agents for a role or workflow into a single install.
Figure 2 — Every connector you add widens what Claude can actually do.
- Google Workspace — Gmail, Calendar, Drive, Docs.
- Microsoft 365 — install the connector, which also embeds Claude directly inside Word, Excel, PowerPoint, and Outlook as an add-in.
- Add others as needed: CRM, project tools, data warehouses, design and document tools. Type a request that needs an app and Claude will suggest the right connector or plugin.
Connect only the accounts you are comfortable letting Claude read and act on. You can review and remove connectors anytime in settings. Use a test folder or account first if you want to build trust before connecting production systems.
Cowork also lets you set how much it checks with you before acting: Manually approve (asks every time), Automatically approve (reviews each action for safety, asks less often), or Skip (no checks, so use it only when you trust everything involved).
Two more worth knowing: web search is on by default (org owners can restrict it), and skills or plugins from outside Anthropic carry real risk, including prompt injection and data exfiltration. Only install what you or your org trusts, and skim what a skill does before you turn it on.
6 Chat vs. Cowork: When to Use Which
| Chat | Cowork | |
|---|---|---|
| Use it for | Quick questions, drafting, brainstorming, one-off answers | Multi-step projects, building files, automating workflows across your tools |
| Works with files | You paste or upload content | Reads and writes real files in a folder on your computer, or its own workspace in a cloud session |
| Runs tasks | Responds in the moment | Executes longer tasks. Cloud sessions keep working on Anthropic's servers even if your computer is off. Use Dispatch when a task needs your desktop specifically |
| Available on | Desktop, web, mobile | Desktop (full capability, including local files and computer use). Also in beta on web and mobile for lighter-weight tasks |
| Think of it as | A fast, knowledgeable assistant | A teammate who gets work done end to end |
7 Get the Most from Chat
- Use Chat for quick questions, drafting, and thinking out loud.
- Chat now remembers context across sessions on its own. But for a project with a lot of background, pasting in (or attaching) your .md context file is still faster than hoping memory caught everything.
- Iterate. Your first prompt rarely produces the best answer; refine with follow-ups like "tighten this," "add sources," or "make it a table."
- Try Learning mode when you want to understand something, not just get an answer. Pick "Learning" from the style dropdown in the chat box and Claude switches to a Socratic approach: it asks questions and walks you through the reasoning instead of handing over the answer. Good for ramping up on a new framework or regulation before you brief the team on it.
8 Artifacts: Build Things That Stay Live
Not every output has to be a static file. An artifact is a live page Claude builds inline (a dashboard, tracker, calculator, or small tool) that you can reopen later.
- Ask for one directly: "build this as a live artifact" or "make this a dashboard I can reopen."
- Live artifacts can pull fresh data from your connectors each time you open them. Useful for anything you would otherwise rebuild from scratch every week.
- On Team and Enterprise plans, you can share a live artifact with colleagues; on other plans it is yours to revisit.
9 Cowork: Your Project Powerhouse
Cowork coordinates work and builds deliverables end to end: documents, spreadsheets, decks, analyses, and multi-step workflows across your connected tools.
Figure 3 — How a Cowork task runs, start to finish.
- Work in projects, not one-offs. Put related work into a project so context and files stay together, and you do not lose track. Each project keeps its own dedicated memory, separate from your other projects.
- Give it real tasks. Be specific about the outcome and the format. "Build a 1-page market summary as a Word doc with a sources table" beats "tell me about the market."
- Use voice mode. Speaking your instructions is faster than typing, especially for longer tasks.
- Let it use files. On desktop, Cowork reads and writes to a folder on your computer, so outputs land as real files you can open and share. Cloud sessions on web or mobile keep their own workspace and hand you finished files to download.
- Schedule recurring work. Type "/schedule" in any Cowork task, or open the Scheduled tab, to have Claude run something automatically, daily, weekly, or on demand, without you kicking it off each time.
- Send it from your phone with Dispatch. Message Claude from your phone and have it work on your desktop, using your local files, browser, and apps, then report back. Keep your computer on and the desktop app open; for work that needs to run with your computer off, start a cloud session instead.
10 Skills: Prebuilt Expertise on Demand
Skills are packaged capabilities Claude can apply to a task, from building a polished Word document to running a specialized workflow. They are one of the most powerful features in the app.
- Type a forward slash "/" to see and call available skills.
- Auditrol already has a few built. The intelligence library tracks regulatory requirements, RFPs, and prospects. The product brief generator builds a customized one-pager for any bank or credit union you are prospecting. Start with one of those instead of building from scratch.
- Skills, connectors, and plugins now share one directory. A plugin packages multiple skills and connectors together for a role or workflow in a single install.
- Skills cover formats (documents, spreadsheets, presentations, PDFs) and full workflows (research, sales prep, financial analysis, and more).
- You can install skills others have built and create your own for repeatable tasks unique to your work. On Team and Enterprise plans, skills can be shared with colleagues or provisioned org-wide.
- When you find yourself giving Claude the same multi-step instructions repeatedly, that is a candidate to turn into a skill.
11 Prompting That Gets Better Results
- Give context up front. Who it is for, the goal, and any constraints. Load your memory file when relevant.
- Specify the output. Length, format, and tone: "one page," "as a table," "executive tone, no jargon."
- Show an example. Paste a sample of the style or structure you want it to match.
- Ask it to plan first. For complex work, "outline your approach before you start" catches misunderstandings early.
- Iterate, do not restart. Refine the existing answer rather than re-asking from scratch.
- Ask for sources. Request links and verify them, especially for anything client- or decision-facing.
12 Common Mistakes to Avoid
Treating it like search
The value is in delegation and iteration, not single questions.
Skipping context
Claude remembers a lot on its own now, but a memory file still beats generic output for anything with real background.
Working one-off in Cowork
Without projects you lose track of files and history.
Defaulting to Opus or Fable
They are powerful but burn usage fast. Sonnet 5 handles most work.
Not verifying facts
Always check sources for anything that carries weight.
Treating the output as final
Claude's draft is a draft. Anything headed to a client, examiner, or regulator gets a human review pass first.
13 Keep Learning
- Anthropic Learning Center: anthropic.skilljar.com. Free, structured courses; mirrors the Anthropic Partner Network onboarding.
- Prompting guide: docs.claude.com, prompt engineering section.
- Build your memory file, learn to call skills, and connect one new tool each week. Momentum compounds quickly.
- If Claude gets something wrong, use the thumbs-down button in the app to flag it. That helps Anthropic, and it keeps your own skills and prompts improving instead of quietly working around the problem.
Your day-one path to value:
- Desktop app
- The full Claude experience on your computer; the only surface with direct local file access, browser control, and computer use.
- Cowork
- The agentic workspace that builds files and runs multi-step tasks across your tools. Available on desktop, and in beta on web and mobile.
- Chat
- Fast, conversational Q&A and drafting, with automatic memory across sessions.
- Cloud session
- A Cowork session that runs on Anthropic's servers and keeps working even when your computer is off or asleep.
- Artifact
- A live page Claude builds inline (a document, dashboard, or small tool) that you can reopen later. Some refresh with current data automatically.
- Dispatch
- Lets you send tasks to Claude from your phone that run on your desktop, using your local files, browser, and apps. Requires your computer on and the desktop app open.
- MCP connector
- A secure link between Claude and an outside tool (Gmail, Microsoft 365, a CRM, etc.).
- Memory file (.md)
- A Markdown document of your context and preferences that you load to ground Claude.
- Skill
- A packaged capability you call with "/" to apply prebuilt expertise to a task.
- Project
- A container in Cowork that keeps related work, files, context, and its own memory together.
- Plugin
- A bundle of skills, connectors, and sub-agents for a role or workflow, installed from a directory in one step.
Which tool, when
Chat
A quick question, a draft, or thinking out loud.
Cowork
Multi-step work that must produce real files.
Dispatch or cloud
You are away from your desk, the work should keep going.