Claude Merged Chat and Cowork Into One Project. The Mode Picker Is Gone

Every cohort I run, someone asks a version of the same question on Day 1: "so do I chat with Claude about this, or do I hand it off as a Cowork task?" I used to have a real answer, because until this week that was a real fork. Chat lived in one place. Cowork, with its files and its scheduled tasks and its multi-step runs, lived in another. You picked a lane before you'd typed a word, and picking wrong meant starting over in the other lane with the same context re-explained from scratch.

That fork is gone. Anthropic merged Chat and Cowork into one Home on claude.ai, on both web and desktop, rolling out in the first week of August 2026 (the launch promo, doubled Cowork usage limits, ran through August 5, which is roughly when this landed for most accounts). One composer starts either a quick back-and-forth or a multi-step task. You stop deciding up front, because there's nothing left to decide.

What actually changed

The visible part is small: a single "New" button, one text box, and a Home sidebar that lists your chats and your Cowork tasks together instead of in two separate tabs. Type a question, get a conversation. Ask for something that needs files, tools, and several steps, and Claude runs it as a task without you having to file it under a different mode first.

The part that matters is underneath. A Project on claude.ai now carries the same four things regardless of whether you're chatting inside it or handing it a task:

Before this week, that last row lived on the Cowork side of the fence only. Set up a project to hold your onboarding docs, and a scheduled task attached to it would read those docs. But a plain chat window on the other side of the app had no idea the project, or its instructions, existed. You'd paste the same context in again, or worse, not bother and get a generic answer. Now a chat started inside a project and a scheduled task attached to the same project draw from the identical memory, instructions, and files. Switching modes stops costing you anything.

Anthropic's own framing, straight from the update: handing Claude a task now "starts the same way a conversation does." That's the whole design decision in one sentence. The mode isn't a container you choose. It's just how far a given exchange happens to go.

The rollout had a real bug, already fixed

Worth knowing if you poked at this in the first few days: chats started inside a Project weren't reliably picking up that project's Instructions and Context links, including the actual files linked there. You'd be in the right project, ask a question, and get an answer that ignored the document sitting right there in the sidebar. That's fixed as of the same release wave. If you tried a merged project on day one and it felt like the project context wasn't landing, that was real, not you misconfiguring something, and it's worth trying again now.

What I'm doing differently

I'm consolidating projects I'd split for no good reason. I had a "Claude Camp content" project for chatting through blog ideas and a separate "Claude Camp ops" Cowork setup for scheduled admin checks, mostly because the two modes used to force that split. There's no reason for two anymore if they'd both benefit from the same instructions and files. One project, one memory.

I'm putting the instructions Claude actually needs into the project, not into every prompt. The whole point of a shared project is that you stop re-explaining yourself. If you've been pasting the same three sentences of context at the top of every chat, that's now a one-time Instructions field instead of a recurring tax.

I'm treating "schedule this" as a natural extension of a conversation, not a separate app. You can turn a chat you just had into a Skill, or ask it to run again on its own, on the same screen you had the conversation. That's a genuinely different habit than opening a different tool to set up an automation after the fact, and it's the kind of thing that only sticks if the friction to try it is close to zero.

If you're teaching yourself the Claude stack right now, or you're the person at your company who's supposed to know how this stuff fits together, this is worth ten minutes today: open claude.ai, look at your existing projects, and check whether the instructions and files you'd want a scheduled task to use are actually sitting in the same project as the chats you have about that work. A week ago that alignment didn't matter. Now it's the whole mechanism.

Sources: Claude Cowork on web and mobile · Get started with Claude Cowork · Claude release notes