Skip to Content

Claude Code gets projects, side chats and split view

Anthropic is rebuilding the Claude Code desktop app, with projects, sub-sessions, side chats and split view.
October 28, 2026

Anthropic is rebuilding the Claude Code desktop app, with projects with sub-sessions, side chats and split view for several tasks at once.

Why this is useful to know

Anthropic has given the Claude Code desktop app a major overhaul, as Ben from BenAI shows in a video. You can put several tasks side by side on one screen, a main chat can have sub-sessions work on parts of a larger project, and you can ask a question without interrupting the work in progress. It is also easier to see how full your conversation is. For most small and medium-sized businesses this is not yet everyday fare, because the features are rolling out step by step and differ per plan. But it shows where AI tools are heading: from a single chat window to a workspace in which several AI agents handle a project together. Above all, remember that the available features differ per plan.

When you need this

Sound familiar? You are working on a larger project, such as a quote, a campaign or a migration, and you have three separate chats open that know nothing about each other. Then a project with a coordinating chat can help. It starts sub-sessions for separate parts and keeps the overview. Note: on a Pro or Max plan you get projects, on a team plan you see sub-sessions. Check first which one you have. Keep one chat per task, and put a question unrelated to the work in progress in a side chat, so your main conversation stays clean. Use the context counter to see when a conversation gets too full, and then have it summarised. Test the features on a small task first before you use them for client work.

How it works under the hood

Technically, it is a coordinating agent that creates sub-sessions (threads), each with its own context window. In projects, the threads share a memory that they contribute to and draw from. In sub-sessions, the sub-session only gets a briefing and a prompt from the main chat, without shared memory. Project threads can keep running in the cloud, sub-sessions only run locally. That matters for data processing: files then stay on your machine, but model traffic still goes to the vendor. Side chats read the main chat up to the moment they are opened and can then send messages back to the main chat, so they do touch the context. Compacting is a harness mechanism that summarises the context, not a cosmetic button. According to the demo, the /doctor command reads your CLAUDE.md, skills and settings and removes outdated or conflicting instructions. That changes behaviour, so test it on a copy first. Verify model names and availability in the official documentation.

This news touches chapters 5 and 23 of the course AI(l) or nothing by Newway. If you follow the course, this subject returns there at a calmer pace.

Written in response to a publication by Ben AI (YouTube), "Claude 2.0 is Insane (All Updates Explained)": https://www.youtube.com/watch?v=J5EbL1UaqG8