Big Idea

The argument in one line.

The rebuilt Claude desktop app turns Claude Code into a multitasking workspace you can reshape yourself, so non-coders can run many long jobs at once and keep them going while away.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You use Claude chat or Cowork every day and want to know what changed in the rebuilt desktop app before you click around blind.
You've avoided Claude Code because the word 'code' sounds like it's only for developers, and you want proof you can use it without reading a line of code.
You run several Claude sessions at once and lose track of which one is finished, waiting on approval or still working.
You want Claude to keep working on long tasks while your laptop sleeps or you're away from your desk.

SKIP IF…

You only use the Claude API or a terminal-based Claude Code setup and never open the desktop app.
You want a deep tutorial on any single feature. This is a fast survey that names each feature and what it's for.

TL;DR

The full version, fast.

Anthropic rebuilt the Claude desktop app so it has two tabs, Home (chat plus Cowork) and Code, and the video calls it a Claude 2.0. New artifact types (documents, slides, design, live dashboards, motion) work in chat and code. Code gets a session rail showing context use, sections for organizing chats, side chats, forks and split view. Remote control, keep-awake and cloud sessions keep work running while you're away. Nested sessions, master sessions and projects handle multitasking, and mods let you rebuild the app's interface by describing the change.

Free for members

Chat with this breakdown — free.

Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.

Create a free account →

Chapters

Where the time goes.

00:00 – 01:40

01 · What changed and why Code matters

Cowork merges into chat, leaving Home and Code tabs. Code is where Claude's power is, and you don't need to read code to use it.

01:41 – 03:32

02 · New artifact formats

Documents, slides, design, dashboards and motion now live in artifacts that work in both chat and code.

03:34 – 05:20

03 · Session rail and new layout

The rail shows location, remote control and context usage, with terminal, changes, browser and files panes beside it.

05:20 – 06:52

04 · Sections, side chats, forks and split view

Organize chats into sections, ask quick questions in a side chat, fork to explore, and view sessions side by side.

06:52 – 09:08

05 · Keep Claude running while you're away

Remote control, keep-awake, cloud sessions and a verifiable goal let work continue without you at the computer.

09:11 – 12:30

06 · Nested sessions, master session and projects

Three ways to multitask: nested sessions for side jobs, a master session to manage running work, and projects for ongoing cloud work.

12:34 – 14:23

07 · Mods

Describe a change and Claude modifies your own app, with examples like a prompt cache keeper, a destructive-action guard and a recording mode.

14:23 – 14:45

08 · Verdict

The combined changes, especially mods and multitasking, make it feel like a new and better Claude.

Atomic Insights

Lines worth screenshotting.

Cowork has been merged into chat, so the Claude desktop app now has only two tabs: Home for chat and Cowork, and Code.
You don't need to read any code to use Claude Code; you chat in plain language with the more capable version of Claude.
Design now lives inside Artifacts, and the standalone design platform is shutting down.
Dashboard artifacts can be live, pulling fresh data from your connectors and updating when your data changes.
Claude's motion graphics are generated through code rather than a video generator, which makes specific tweaks fast and precise.
The session rail shows how much of the 1 million token context window is used and which connectors, skills, agents and memory files are eating it.
A side chat can read the whole main chat and answer a quick question without adding anything to the permanent conversation.
Forking copies a chat from a chosen point into a new chat, so you can try another direction without touching the original.
The keep-awake setting lets your computer sleep normally once Claude finishes, instead of staying on all night.
A goal only works if the finish line is something Claude can verify. 'Make it great' is too vague to check.
Nested sessions run each piece of a big job as its own chat under the main one, so you can follow up with one piece directly.
A master session is a technique, not a button: pin one chat and ask it what's waiting on you across your other sessions.
A master session sees your 20 most recent local sessions by default and can't see sessions running in the cloud.
Projects differ from nested sessions because their threads run in the cloud by default and start with the project's shared instructions and memory.
A prompt-cache mod sends a tiny message before the one-hour cache expires, so returning to an old chat doesn't force a full reread.

Takeaway

Claude's desktop app is now a multitasking workspace.

WHAT TO LEARN

The rebuilt app rewards you for moving from single chats to organized, parallel, long-running sessions that you can steer from anywhere and reshape with mods.

02New artifact formats

Artifacts now cover documents, slides, design, live dashboards and motion, and they work in both chat and code, so most people can use them right away.
Live dashboards pull fresh data from connectors, so a single artifact can stay current instead of being regenerated by hand every time.

03Session rail and new layout

The session rail's context view shows what is consuming your 1 million token window, so you can audit connectors, skills and memory files before you compact.
Use the context ring beside the model picker for an at-a-glance read on how full the window is without opening the rail.

04Sections, side chats, forks and split view

Sections group chats by project or topic, side chats answer quick questions without polluting the main thread, and forks let you test another direction safely.
Open related sessions in split view instead of hopping through the sidebar to see which one finished first.

05Keep Claude running while you're away

Turn on remote control in settings so every new session on that computer can be approved and answered from the Claude app on your phone.
Enable keep-awake so Claude holds your computer awake only while it works, then lets it sleep, rather than leaving it on all night.
Pair a goal with a finish line Claude can verify, such as a passing check, so it keeps iterating on its own instead of needing follow-up prompts.

06Nested sessions, master session and projects

Use nested sessions for side jobs like researching several competitors at once, then ask follow-ups to one child chat without clogging the main one.
Pin one chat as a master session and ask what's waiting on you. It sees your 20 most recent local sessions and can start, rename, sort and message them.
Choose projects when threads should share instructions and memory and keep running in the cloud with your laptop closed. Projects are still in beta.

07Mods

Mods let you change the app by describing the change, from side panels and status bars to notifications, with no code needed from you.
A prompt-cache keep-warm mod sends a tiny message before the one-hour cache expires, which costs far less than rereading a long chat from scratch.

Glossary

Terms worth knowing.

Artifact: A standalone, shareable output Claude creates in its own window, such as a document, slide deck, design, dashboard or motion graphic, instead of a plain text reply.
Session rail: A panel on the right of a Code session showing where it runs, its folder, the remote control toggle and how the context window is being used.
Context window: Everything Claude holds in mind at once during a chat. When it fills, older details get summarized, which is called compacting.
Remote control: A setting that lets you see a running session and reply or approve actions from the Claude app on your phone.
Nested session: A sub-chat that appears under a main chat, runs a piece of a larger job in parallel and reports back.
Master session: A pinned chat you use as a control center to check on, start, rename and message your other sessions.
Prompt cache: A saved copy of the conversation Claude already read, so later messages cost fewer tokens. It goes cold after about an hour.
Mod: A customization of your own Claude app, built by describing it to Claude, such as a side panel, status bar or notification.

Resources

Things they pointed at.

01:16productClaude 2.0 setup kit (checklist, interactive tour, prompts, mod starter pack)

12:30toolPrompt cache keep-warm mod

13:38toolBlast Radius mod

14:00toolRecording mode mod

Quotables

Lines you could clip.

00:51

“You don't need to look at a single line of code to use it.”

Short, reassuring, removes the biggest barrier to trying Claude Code.→ TikTok hook↗ Tweet quote

08:48

“Just make sure you write a finish line that Claude can actually check. Don't use something vague like make it great.”

A concrete rule for goals that fits in one sentence.→ newsletter pull-quote↗ Tweet quote

The Script

Word for word.

Read-along

Don't just watch it. Burn it in.

See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.

Anthropic has completely rebuilt the Clod desktop app. It has new powerful features and is better in a bunch of different ways. It's a big enough change that one of the engineers who worked on Clod code said it's basically a Clod 2 .0.

There's new artifacts, a new layout, split view, new ways to multitask, mods, and a lot more. I'll cover everything that's new and how to get the most out of it.

There's been changes all throughout the desktop app. A big one is that there's no more co -work tab. That has been merged into the chat tab.

So now there's only two tabs, home, that includes chat and co -work, then code. To get the most out of Clod, you do need to be using code. That's where you'll unlock all of its power.

And if you've never used it, don't be scared away by the word code. You don't need to look at a single line of code to use it. I've wondered if they'll eventually change the name of it at some point.

It feels misnamed since so many people who use it, myself included, don't know any code and don't even see the code that comes out of it. You just chat in natural language like you do in chat, but while using the far more capable version of Claude. And a lot of these new features make it even easier to use for newcomers than it already was.

Now, I did also put together a free Claude 2 .0 setup kit that goes with this video. It has a setup checklist, an interactive tour of the new layout, prompts for everything I show, and a starter pack of mods that I've tested. I'll get to those later, but you can modify Claude's interface or functionality any way you want to customize it to your own workflow.

That resource is free, linked in the description.

And before those code features, I'll start with artifacts, since these work in both chat and code. So anyone can use them right away. There's some extremely useful new options in here.

And an artifact is when Claude makes something substantial that needs a standalone home. When you're chatting, it gets a dedicated window. It isn't just a text response.

And these are shareable, so you can show them to your team, present, download, or do all sorts of things with them. The new formats they have are Documents, Slides, Design, Dashboard and Motion. The first two are fairly straightforward.

Docs is like a Google Docs you can chat with. Slides you can create and present from Clod or download as a PowerPoint. Design also has its own dedicated platform, but that is shutting down now that they've moved it into Artifacts.

This one is extremely powerful for designing prototypes, social posts, websites, apps, or any visual work. I have done multiple full videos just on design. Dashboards make charts, numbers, tables from your sources.

But these new dashboards can be live, pulling in fresh data from your connectors. Then they'll update when your data changes. You can connect them to your analytics, emails, updates, revenue, and things like that.

Then there's motion. This is how you can make all sorts of motion graphics and animations. Cloud makes incredible motion graphics, and it does it all through code.

It doesn't have a normal AI video generator. But since it's done all through code, it's really easy and fast to make specific tweaks to the output. I use Clod for all the motion graphics in my videos, including this one.

Although I use it through a full video editing platform I built with Clod code. These new artifacts may not all be available to you yet. Dashboard and motion are still in beta.

I have an Ultra plan and don't have the motion option available yet, actually. But they'll be there for at least all the paid plans over time.

Now I'm going to move on to some of the new Cloud Code features. This is where things will be the most powerful. I'll start with some big changes to the interface that make it a lot easier to understand everything that's going on.

Then we'll get into the more function -based stuff. On the top right, you can open up the new session rail. This gives a lot of helpful information.

If it's running on your computer or in the cloud, which folder it's working in, then remote control, you can toggle on and off. That lets you steer that chat from your phone. More on that later.

Then context shows you how much of your 1 million token context window is being used and what's using it. And a context window is everything that Clod is holding in its mind at once. When it fills up, you can compact it or it will automatically.

Then all the older details get summarized. Messages is usually the largest portion, but then there's your connectors, skills, custom agents, memory files, plus dropdowns to see exactly which of those it's using. You can see if something in particular is eating up your context and audit exactly what's going on.

And down below, they do also have a context ring right next to the model picker that gives the same information at a glance without having to open up the rail. And if you want to click through all of this yourself, there's an interactive tour of the whole layout in the setup kit. It'll make everything a lot easier to understand what's going on here.

You can open and close that session rail at any time. Then next to that, you have more options. Some are already here.

There's the terminal if you need it. Changes to show the history of changes it's made. The browser where it can pull up tabs and search the internet and view websites.

Then there's files that show the full list of files it's built or is working with. You may never want to look at those, but for anyone that was used to an IDE like Cursor or VS Code, if you like having that file structure, it's in here now. Then on the sidebar on the other side, this has a new feature that's really helpful called sections.

It's basically folders for organization, but it's really nice. Rather than just a group for everything being worked on within one location on your computer, you can group them into different projects or topics, however you choose. So you click the three dots and hit Move to Section.

Then create a new section. Once you have one, you can drag chats in. This has helped me stay much more organized.

When Claude is in the middle of working on something, you typically don't want to interrupt it, but you may have a separate question you want to ask. You can do that in a side chat. So click the three dots and hit side chat, or use command or control plus.

Then it will open up a chat where you can send off a quick prompt. That can read the entire main chat and respond to you without adding the response to the permanent chat. These side chats aren't saved permanently, so if you do quit, it won't be there when you come back.

So that is different from a forked chat, which is also really useful. But forking copies the chat from that point into a new chat. So you can try a different direction without messing up the original.

There is also a split view. Right click, then hit open in, then split view. Then your sessions will be right next to each other.

I'll typically be working on a few sessions at once. Rather than having to jump between them on the sidebar, then just waiting to see which finishes, then jump back and forth, I'll pull every session I'm working on up in split view. And I'm not even sure what the limit is to how many of these you can have open at once.

I pulled up a ton just to see and didn't hit the limit, so it's definitely more than you would ever want. Just a few is typically all I have open.

There's some new options that make it simpler to keep Claude running while you're not at your computer or while your computer is closed. Remote control is where you can see your session and steer it from your phone. They made it a lot easier to switch this on.

In a session, they have the toggle I showed earlier where you can just switch it on. Or if you go into your settings, there's also a toggle there that says connect new sessions to remote control. Then it will be on automatically for every new session you start on that computer.

Then on your phone, it shows up in the Cloud app under code and you can reply or approve things right there. I use remote control all the time. It's amazing.

And I love having the option to do this for all of them so I don't have to remember to switch it on. Then lower in your settings, you'll see the Keep Computer Awake While Cloud Works option. This is another big quality of life setting.

You could ask Claude to do this before, but it was definitely easy to forget. I'd keep my computer's settings to never go to sleep unless I did it manually for when I go out, or a lot of the time I'll have it working on something big right before I go to bed. Now I can switch this on and leave my computer settings to go to sleep after a certain amount of time.

Then Claude will keep it awake the whole time it's working, but when it's done, my computer will be able to go to sleep instead of staying on all night. That option is available session by session as well, but I leave it on in my main settings, so it's on all the time. You can also run sessions in the cloud, so they'll keep going while your laptop is closed.

When you start a session, you can pick cloud instead of your computer. They won't be able to see the files on your computer, but a lot of the time you'll have a GitHub repo for something you're building. It can just work from that while your computer's off.

These features to keep things running longer and while you're away are all really helpful, especially if you use forward slash goal, which I regularly do. That's where you give Claude the finish line, then it keeps working until it can verify it's completed it. That saves a lot of follow -up prompts, since Claude just checks its own work, then keeps working and changing until it gets there.

but it can run for a long time when you work this way. So having an easy way to keep your computer on is nice. Just make sure if you use this that you write a finish line that Claude can actually check.

Don't use something vague like make it great. So with a goal set and remote control on, I can give it the finish line, walk away, and just check in from my phone.

These next features are all really powerful new ways for Claude to multitask better. There are three different ways Claude now runs more than one session, and therefore different things, although a couple of them seem similar. Nested sessions let Claude split a big job into sub sessions that run in parallel and report back to the main chat.

Claude has used sub agents to run things in parallel for a long time now, but with nested sessions, each piece runs as its own chat nested under the main one. so you can see it and follow up with that chat directly. It could be researching five different competitors all running at once in separate nested sessions.

The main chat combines those results, but you could ask follow ups about one specific competitor without clogging up that main chat. And you can just ask for these, like split this into separate sessions for each competitor, or Cloud will sometimes offer it with a task chip. The task chips are Basically, Cloud will notice related work, then pop up a card.

Then if you hit start, it opens it up as a nested session under the chat. A master session is one pinned chat that acts as your control center for all the others. This isn't a dedicated feature.

There's not like a button for this, but it's a new technique people have been using. That's because every code chat can now see and manage your other chats. So you pick one chat and make it your control center.

Then you can pin that to the top, or you could just make a new section and add that to the top. Then you can have it help orchestrate all your other sessions. If you ask what's waiting on me, it can check all your other sessions and list which are finished or need approvals or which are still working.

It can start, rename, sort, and send messages to the other sessions. Like I have a build that has a bunch of different pieces to it, and those are each in different chats, but I have one master session to interact with, so I don't have to dig through the different chats to find what I was doing. You can even have a master session for each project, then a separate master session for your whole cloud code if you want.

By default, it sees your 20 most recent sessions, and you have to ask explicitly if it needs to look further back. Also, it only sees sessions on your computer, not ones running in the cloud. Now, projects is a new feature in Cloud Code that gets confusing because there's a separate feature called projects in chat, but this one works differently.

So projects also do have a main chat that acts like a team lead and briefs separate threads that each do one piece of the work, similar to nested sessions. There are two big differences. With projects, those threads run in the cloud by default, so they keep going with your laptop closed.

although you can ask for one to run locally. And every new thread starts with the project's instructions and memory, where a nested session only gets what the main chat tells it. Then it also has an overview that shows what's waiting on you, plus a library of every file.

So, projects have Claude as an orchestrator working towards your goal, and it will produce separate threads, in the cloud or locally, that each have the project's context and memory. These are in beta, so you may not see them yet. Again, all of these features will roll out to everyone over time.

But to summarize these, nested sessions are for side jobs, a master session is for managing what's already running, and projects are for ongoing work where every chat shares the same memory and can keep going in the cloud. If you're still not sure which one to use, the setup kit has a quick three -question picker that tells you, plus copy -paste prompts to start each one.

Mods are an awesome new feature where you can modify your own personal Clod app to add features, change the interface, or make it work more to how you want it to. You can change anything just by describing it to Clod. So that could be a new type of side panel or status bar, different pop -up notifications, or things much more elaborate or random, just like Jared Watts on X made one that connects you to a multiplayer Doom server while Clod is busy.

So you just open up a code session and say something like, make me a mod that does this, and it will build it into the app. One I've found really useful is one I saw Nate Hurk make. It keeps your prompt cache warm.

For every message you send, Claude rereads the whole conversation. The cache is a saved copy of what it already read, so rereading uses a lot less tokens. But this goes cold after an hour.

So coming back to an old chat means Claude has to reread the whole thing. But he built a mod that keeps that prompt cache warm. Basically, it sends a tiny message right before the cache expires.

And that costs a lot less than having Claude reread a long chat history from scratch. That one's really useful, so I added it to the mod starter pack that's in my setup kit. There's that one and seven others that I tested.

These aren't ones that I built, but they're mods I either found on GitHub and tested, or official Anthropic mods. There's Blast Radius, which steps in before Claude runs something destructive, like deleting a folder or something, and then it shows you exactly what would be lost with a button to cancel. And recording mode is another great one for me.

It hides things like your API keys or email addresses that are on screen. That way it's hidden while I'm doing a screen recording. It's super easy to switch on and off.

In that guide, each one of these has a prompt you can copy and paste into Clod and that will install it for you. And there is a lot more you can do with mods. You can completely reshape the Clod app around how you work.

So that was a ton of new features. Is this really enough to call it a Claude 2 .0? For how I use it, I would say yes.

The combination of all of these new features is a massive change for me, especially being able to mod it and to multitask better. Plus just the difference in the layout and just countless quality of life improvements. It feels like a whole new and better Claude to me.

The Hook

The bait, then the rug-pull.

Anthropic rebuilt the Claude desktop app so thoroughly that one Claude Code engineer called it basically Claude 2.0. This walkthrough covers every major change in about 15 minutes, from new artifact types to running a fleet of sessions from your phone.

Frameworks

Named ideas worth stealing.

09:11list

Three ways Claude runs multiple sessions

Nested sessions for side jobs
Master session for managing what's already running
Projects for ongoing work with shared memory that can run in the cloud

Nested sessions split one big job into parallel child chats, a master session is a pinned chat that oversees the rest, and projects give every thread shared instructions and memory.

Steal forchoosing the right multitasking setup for a build

Storyboard