The Problem You Have Felt
If you use Claude Code seriously, you know this moment. You fire off a prompt that will take Claude thirty minutes to finish. A refactor across twelve files. A video model benchmark. A data migration dry run. Something real.
Then you sit there.
You cannot start a second task. You cannot switch to a different repository. You cannot even safely change branches in the repo Claude is working in, because Claude's mid-diff state assumes the working tree stays put.
So you wait. Or you open a second VS Code window, authenticate again, configure a second session in a different repo, and now you are context-switching between two projects at once. Neither session knows what the other is doing. Branches collide. Review becomes a mess.
That is the era Anthropic just ended.
What Shipped
The new Claude Code Desktop app is a full redesign around one idea: you are the orchestrator, Claude runs parallel agents. The app's entire layout serves that mental model.
Sidebar with parallel sessions. You can run multiple Claude Code sessions at once. Each one gets its own entry in a left rail. Click a session, the main panel switches to that session's state. Everything keeps running when you click away.
Git worktree isolation per session. This is the one that matters. Every parallel session automatically runs inside its own git worktree. If you open three sessions against the same repository, you get three separate working trees, three separate checkouts, three separate branches Claude can push to without any of them stepping on each other. This was technically possible before, but you had to wire it up by hand. Now it is the default.
Drag-and-drop panel layout. Inside each session you get integrated panels: terminal, file editor, diff viewer, HTML and PDF preview. You drag them wherever you want. Want the diff viewer on the right and the terminal docked bottom? Drag. Want the preview pane to take half the screen while Claude works? Drag. The layout saves per session.
Built-in PR review. Claude's changes surface in a visual diff viewer that is faster than the terminal-based review most of us have been doing. You approve, tweak, or reject inline. When you ship, the app monitors the pull request on GitHub and can auto-merge on green CI if you configure it to.
No more second VS Code window. The terminal, editor, diff review, and preview all live in the same app. You can still use VS Code's Claude Code extension if you want, but the new desktop app stands on its own.
What Stays the Same
Your CLI still works. The VS Code extension still works. Existing sessions are not auto-migrated. If you already have a cron-based workflow running on a VPS with Claude Code CLI inside a tmux session, none of that changes.
The desktop app is additive. You install it from the download page, log in with your Max or Team account, and it shows up alongside whatever you already use. No migration pain, no "upgrade or die" pressure.
Install
- ●Go to the Claude Code download page on claude.com
- ●Download for Mac, Windows, or Linux
- ●Install, log in with your Pro, Max, Team, or Enterprise subscription
- ●It runs next to VS Code. Use whichever fits the task.
Get the Weekly IT + AI Roundup
What changed this week in NinjaOne, ServiceNow, CrowdStrike, and AI. One email, every Monday.
No spam, unsubscribe anytime. Privacy Policy
First run: open three sessions against your three most active repos. Leave them idle. Now you have parallel channels ready to go for the next time you want to fire off two unrelated tasks at the same time.
The Mental Model Shift
The interesting thing about the redesign is not the features. It is what the features imply. Anthropic has stopped treating Claude Code as a terminal tool with an AI attached and started treating it as a workspace where you manage several AI agents at once.
For anyone juggling multiple active projects, that shift matches the shape of the work. A single orchestrator surface with several parallel channels handles reality better than a stack of disconnected terminals ever did.
Where This Leaves You
If you are running one Claude Code session at a time and shipping fine, the old setup is still valid. The desktop app does not make existing workflows better, it just makes new workflows possible.
If you are running multiple projects, hitting long tasks often, or reviewing pull requests from yourself and from Claude at the same time, the upgrade is obvious. Install it. Open three sessions. Let your work spread out.
Claude Desktop vs VS Code Claude Code: Which to Use
Both surfaces support multiple Claude sessions now. They are not the same thing. Here is the honest comparison after using both for real work.
| Capability | Claude Desktop | VS Code Claude Code |
|---|---|---|
| Multiple sessions in one window | Yes, sidebar | Yes, tabs |
| Multiple OS-level windows | No, single app | Yes, open as many as you want |
| True process isolation between sessions | No, one process | Yes, one process per window |
| Crash blast radius | All sessions die | Only that window dies |
| Git worktree per session | Yes, automatic | Manual setup |
| Drag-and-drop panels (terminal, diff, preview) | Yes | No, stuck with VS Code's panel layout |
| Built-in PR review and auto-merge | Yes | No, use GitHub or gh CLI |
| Routines (scheduled tasks while laptop sleeps) | Yes, desktop only | No |
| Code editor experience | Basic | Full VS Code, extensions, debugger |
| Shared notification and permission queue across sessions | Yes, can pile up | Per-window, cleaner separation |
Pick Claude Desktop if: you want the orchestrator workspace with worktree isolation handled for you, you care about Routines, you are reviewing PRs all day, or you just prefer the chat-app feel over the IDE feel.
Pick VS Code Claude Code if: you live in the editor anyway, you want hard isolation between sessions, you need the full VS Code extension ecosystem, or you want one window per project so a crash in one repo cannot take down your other work.
Pick both if: you are not weird about it. They share auth and rate limits but otherwise stay out of each other's way. A common shape we have landed on is Desktop for orchestrating long parallel runs and PR review, VS Code for deep focused editing on whichever repo needs real attention right now.
A Related Problem Worth Solving
Parallel Claude sessions are great once you know WHICH Claude session to run for WHICH task. Picking the right model for code review, the right model for long-context refactor, the right model for quick fixes. That is a different kind of matching problem.
LLM Match Maker was built for exactly that. We match people to the AI tools that fit their actual workflow, not whichever one has the loudest launch. If you are standing up a parallel Claude Code desktop and wondering which other AIs to pair it with, we have the answer.
The desktop app ends the "one session at a time" era. Picking the right AI for each session is the next choice. That is where we come in.