Claude Code · Change note
Claude Code’s Background Computer Use Makes Room for My Own Work
Claude Code can work in background windows on supported Macs while I keep using my computer. The change makes delegation feel more comfortable, with important limits around permissions, the local machine, and what background operation actually means.
By Raşit Akyol ·
The change I care about in Claude Code is practical: handing over a task no longer makes my computer feel temporarily unavailable. In my earlier experience, computer use took over the screen. With background operation, I can continue with other work while Claude operates in the applications I have allowed. That feels closer to the delegation experience I associate with Codex.
This is my usability impression from September 1–5, 2026. It is not a timed comparison between the two products. The question that matters to me is whether giving an agent a task also means giving up my own working space.
What changed, and where
Anthropic documents background windows as the default for computer use on macOS 15 or later. Claude can interact without taking over the pointer and keyboard, generally waiting while the user types. It asks for permission the first time a task needs the full screen in each session. Computer use is a beta feature for Pro and Max; these documents exclude Team and Enterprise.
The official announcement says people who previously enabled computer use already have it switched on. Otherwise, the entry point is Settings → General → Computer use. The announcement covers Cowork and Claude Code on macOS. In this note, Claude Code means the Code tab in Claude Desktop, not the terminal CLI.
Why this changes the handoff
Previously, the agent’s work and my own work felt sequential: one of us could use the screen comfortably at a time. The new behavior makes them feel more compatible. The improvement I am describing is reduced interference, rather than an unmeasured claim that Claude completes the underlying task faster.
Claude’s desktop documentation distinguishes browser automation, connectors, shell work, and computer use. A successful browser task therefore is not automatically a test of background interaction with a desktop application. That distinction helps explain why two people can describe different experiences while both are using Claude.
What other users add
In the official announcement discussion, a commenter describes the earlier foreground behavior as frustrating when trying to work at the same time. That matches the problem in my own note. It does not establish how reliably the new implementation works: the discussion is anecdotal, with no repeatable before-and-after test.
Where I would draw the line
Background operation still requires the local computer to stay awake and Claude Desktop to remain open. App permissions still apply. It does not mean the same task keeps running on a separate cloud machine after the laptop is shut.
For me, this is a welcome improvement in how Claude shares a computer with its user. A useful next test would record one specific task, the build and permissions, interruptions during simultaneous work, and whether the final result is correct. That test has not been performed for this note.
Limitations
Qualitative experience from September 1–5, 2026; exact build, plan and task were not supplied. No timed comparison or public original test artifact. The documented background feature applies to macOS 15 and later; this note does not establish Windows support or a reliability rate.