Lesson 04
How you'll work with Claude on your desktop
What you'll learn
Estimated time: 6 minutes
By the end of this lesson you'll be able to:
- Distinguish the ways you work with Claude on the desktop — working with Claude turn by turn, handing whole tasks off for Claude to run, and building software in your codebase
- Recognize which shape of work a task calls for before you start it
- Find where each way of working lives in the desktop app today
Working with Claude on your desktop
The Claude desktop app is your home base for working with Claude — from a quick question mid-meeting to a report Claude assembles from six sources while you do something else. The work sorts into three shapes, and knowing which one you're in is the whole skill of this lesson:
- Working with Claude, turn by turn. You and Claude go back and forth. You ask, Claude answers, you steer, it revises. The thinking happens in the exchange.
- Handing work off to Claude. You describe an outcome — a finished brief, a formatted deliverable, a task that runs every Monday — and Claude plans it, does it, and comes back with the result. You review the plan and the output; you don't stitch the steps together yourself.
- Building software with Claude Code. Claude works directly in a codebase: reading it, writing and testing code, running commands. Built for developers, and worth knowing about even if you never open it.
The first two are where most knowledge workers spend their day; the third is the developer's workspace. The rest of this lesson makes the difference concrete, then shows where each one lives in the app today.
In the product today. Turn-by-turn work happens in Chat. Work you hand off runs in Cowork. Building software happens in the Code tab. All three live in the Claude desktop app.
Working with Claude, turn by turn
This is Claude as a thinking partner: the shape of work where the value is in the exchange itself. You bring a half-formed idea, an unfamiliar dashboard, a paragraph that isn't landing — and you work it out together, one turn at a time.
Reach for this when:
- The answer changes what you ask next. You're brainstorming, and each response opens the next question. You couldn't have written the whole request up front, because you didn't know yet.
- You want to stay in it. Drafting, editing, thinking out loud — the point is your judgment on every turn, not a finished thing at the end.
- It's quick. A question, a rewrite, a "what does this mean?" — small enough that setting up a whole task would be overhead.
Try it out when:
- You're staring at an unfamiliar dashboard. Screenshot it and ask "what do these metrics mean?" Claude explains while the dashboard stays in view, and your follow-up ("okay, which of these should I actually worry about?") is the next turn.
- You're between meetings and need to structure a presentation. Talk it through by voice; Claude drafts an outline from what you said; you push back on section three; it revises. Four turns, done before your next call.
- You've been jotting product-launch ideas across Apple Notes for weeks. Ask Claude to pull together everything about the launch, figure out what you left half-finished, and check your other connected tools for gaps. Then work the gaps together.
In the product today. In the desktop app this is Chat — the same Claude you know from claude.ai, plus a few things that come from running natively on your computer:
- Quick entry. Double-tap the Option key (Mac) to pull Claude up over whatever you're working on. It answers in a compact window that stays on top as you switch apps.
- Screenshots and window sharing. Capture a screenshot or share a window so Claude sees what you see. (Mac)
- Dictation. Talk through a problem instead of typing. (Mac)
- Desktop connectors. Connect local tools and services so Claude can work with what's on your machine.
Handing work off to Claude
Working agentically with Claude is a new way of working for many people. Instead of asking a question, you hand Claude the whole piece of work — gather the context, do the analysis, produce the finished thing — and it comes back done. You're delegating, not just chatting.
Reach for this when:
- The task has several steps you'd normally do in sequence. Pull the figures, compare them, draft the summary, format the doc. Handed off, that's one instruction, not four errands.
- The output is a real deliverable. A Word doc, a spreadsheet, a deck, a formatted PDF — saved where you need it, not pasted into a chat window for you to reassemble.
- The work spans your tools. Meeting notes in one place, the thread in Slack, last quarter's numbers in a spreadsheet. Set up a Friday roll-up as a scheduled task and Claude gathers all three itself every time it runs — nothing for you to round up first.
- It should happen on a schedule, or while you're doing something else. A Friday review of what shipped. A Monday briefing that preps you for your next meeting.
Handing work off doesn't mean stepping back from it. Before Claude starts, it may ask a few questions to pin down scope and format, and it shows you the plan. As it works, you can watch the task take shape — the sources it's drawing from, the files forming, its progress through the plan — and steer at any point. And when Claude is set to ask before acting, it stops for your approval on the actions that matter, like sending an email or sharing a file. You stay in control of what leaves your desk.
Try it out when:
- You want to query all your tools like a database. "Review what we decided about pricing last quarter across meeting notes, Slack, and email, then update the Q3 deck with the findings." Claude finds the answer across all of them and updates the deck. Hand it off, keep working, check the result.
- You have a folder of 50+ project documents — contracts, financial reports, meeting transcripts. Ask Claude to find the ones most relevant to your initiative and produce a summary memo. It reads every page and pulls out the patterns that only emerge from reading all of them. Review fifty like you'd review five.
- You do the same work every Monday morning — check messages, assemble a status update, prep for the day's meetings. Set it up once as a scheduled task, and start Monday with answers instead of admin.
In the product today. In the desktop app, you can hand off tasks in Cowork. What that gives you today:
- Local folder access. Point Claude at a folder; it reads what's there and saves finished work back to the same place. This is the concrete difference from turn-by-turn Chat, which can read what you upload but hands finished files back as downloads rather than saving them into your folder.
- Scheduled tasks. Set a task once — a daily briefing, a weekly roundup, a morning inbox triage — and Claude runs it on the cadence you set. If your computer or the app was closed at the scheduled time, it catches up when you're back.
- Subagents. For a big job, Claude splits the work across background workers running in parallel, each with its own context, and hands you one finished deliverable.
- Projects. Group related tasks into a workspace with its own files, instructions, and memory — like projects in Chat, but built around the tasks you run.
- Browser use. With Claude in Chrome, Claude navigates websites and pulls what it finds straight into the task — competitor pricing across ten sites, data from pages with no API.
- Computer use. When there's no connector for what you need, Claude can operate your computer directly — clicking, typing, opening apps — asking permission before each app it touches, with a blocklist for anything off-limits. In research preview on Pro and Max plans.
- Plugins. Ready-made bundles of skills, connectors, and agents built for a specific kind of work — a sales plugin, a finance one, a legal one — so Claude works the way that role works. Browse and add them under Customize → Plugins.
Cowork is available to Pro, Max, Team, and Enterprise users, with new capabilities added regularly.
Building software with Claude Code
If you write code, the desktop app gives you a full development environment. Claude works directly in your codebase — reading what's there, writing and modifying code, running commands. Visual diffs show what changed, a built-in terminal shows commands as they run, and git tracks every version so you can always roll back. If you're not a developer, the takeaway is just this: it's a separate tab, and this course doesn't need it — Claude Code in Action covers it in depth.
You choose where the work happens:
- Local. Select a folder on your computer and Claude works directly with those files — reading your project, using local tools, and running a development server you can preview in your browser.
- Cloud. Connect a GitHub repository and Claude works in a cloud environment. Sessions continue even if you close the app, so you can start a big refactor and check back later. Good for larger codebases, or when you want to keep the work off your machine. You can run Cowork in the cloud in the same spirit — see Cowork in the cloud in the Introduction to Claude Cowork course (in beta, on eligible plans).
You also choose how much Claude does on its own, with settings including:
- Manually approve. Claude proposes every change and waits for your approval.
- Accept edits. Claude applies file edits automatically.
- Plan. Claude creates a plan before making changes.
In the product today. This lives in the Code tab of the desktop app, available on Pro, Max, Team, and Enterprise plans. You can run multiple sessions across projects and filter them by environment (Local or Cloud) and status from the sidebar.
Choosing the right shape for the task
You won't pick a tab first — you'll notice what kind of work is in front of you, and the tab follows. Here's the whole lesson in one table.
| You're about to… | The shape it takes | Where it lives today |
|---|---|---|
| Ask, brainstorm, draft, or think something through, turn by turn | Working with Claude, turn by turn | Chat (quick entry, dictation, screenshots) |
| Hand off a multi-step task that ends in a finished deliverable, spans your tools, or runs on a schedule | Handing work off | Cowork (folder access, connectors, scheduled tasks, subagents) |
| Write, test, run, and ship code in a codebase | Building software | The Code tab (Local or Cloud) |
Lesson reflection
- Think about how you used Claude this week. Which requests were turn-by-turn thinking, and which were really whole tasks you fed in one question at a time because that's the habit?
- Take the task you'd most like off your plate. Is it multi-step, does it end in a real file, does it span your tools? If yes to any, it's a hand-off — write down the outcome you'd describe to Claude, not the first question you'd ask.
What's next
In the next module, you'll learn how to organize your work and knowledge using projects.
Feedback
As you progress through the course, we'd love to hear how you're using the Claude desktop app in your work, plus any feedback you may have. Share your feedback here.