Anthropic has announced a major update to Claude’s computer control capabilities on macOS. Background computer use is now available in Claude Code and Claude Cowork, giving Pro and Max subscribers the ability to let the AI agent click, type, and open apps while they continue working in other applications.
The announcement came on September 2, 2026, via a post on the official Claude X account. “Claude can now use your computer in the background in Claude Cowork and Claude Code,” the post said. “Give it something to do on your desktop and Claude clicks, types, and opens apps just like you would, while you work on something else.”
This update follows a broader industry trend. OpenAI’s Codex-turned-ChatGPT brought background computer use to the Mac earlier in the year. With Anthropic now matching that feature, the concept of an AI assistant that can work alongside a user rather than taking over their screen has become a meaningful part of AI-powered productivity.
What makes background computer use different?
Claude’s ability to operate a computer is not new. Anthropic first demonstrated the idea with a research preview that let developers connect Claude to a desktop environment and ask it to perform tasks. In practice, however, the earlier approach demanded a lot of attention from the human user. When an AI agent took control of the mouse cursor, the user could no longer use the same screen and pointer at the same time. They were forced to sit back and observe, making sure the model did not click on something wrong. For anyone juggling multiple applications, that was a limitation rather than a productivity win.
Background computer use removes that friction. Instead of waiting for the AI to finish, Mac users can split their attention across tasks. After giving Claude an instruction, they can check their email, write a document, or browse the web while the AI works in parallel. This turns a previously serial process into a more natural multitasking experience.
For example, a user might ask Claude to organize downloaded files into proper folders, rename a batch of images, or populate a spreadsheet from a set of receipts. While the AI executes that process, the user can prepare for a meeting or respond to messages. According to Anthropic, Claude can open apps, type text, click buttons, and navigate macOS exactly as a person would, without requiring the user to remain at the keyboard.
Available in Claude Code and Claude Cowork
The feature has been added to two separate environments, each aimed at a different kind of task. Claude Code is Anthropic’s command-line coding agent. It is heavily used by developers who want to delegate programming tasks to an AI, including writing code, running tests, and interacting with version control. Running Claude Code in the background allows developers to hand off a task like creating a new feature or debugging a failing test and then move on to another project or meeting.
Claude Cowork, on the other hand, is designed for broader desktop work. It expands the AI’s potential beyond development, handling everyday tasks involving apps, files, and productivity software. With the new background capability, Cowork can clean up a desktop, process a series of images, or assemble a presentation while its user shifts focus elsewhere. The combination of the two products means both technical and non-technical users stand to benefit from the update.
Anthropic notes that background computer use in these environments works only on macOS and is available to Pro and Max subscribers. That might initially seem restrictive, but Macs have become a central platform for AI agent development. Apple’s ecosystem, with its tight integration between hardware and software, gives AI models a stable environment to operate in. By targeting macOS first, Anthropic can fine-tune the experience before expanding to other operating systems.
Why multitasking matters for AI agents
When AI computer agents first entered the mainstream, the typical experience involved full-screen cooperation. The user would open the AI interface, start a task, and then watch the model take control of the desktop. That was useful for demonstrating what the technology could do, but it was not ideal for real-world productivity. A worker cannot easily abandon their machine for five minutes every time they delegate a task.
Enabling background operation represents a shift from supervised AI to more autonomous AI. Users can give Claude an objective and check in on its progress when ready. This pattern more closely matches the way human assistants work. You do not need to watch someone type a letter or organize a filing cabinet; you ask them to do it and then move on to other responsibilities.
There are also benefits for computing resources. Running a background task can make use of idle time. If a Mac user opens Claude in the background while working in another app, the machine is doing more simultaneously, but the AI can handle monotonous work at whatever pace the hardware allows. For those who often wait for repetitive tasks to be completed manually, the new mode could transform their daily workflow.
Anthropic and OpenAI are competing on agents
Anthropic is not alone in exploring agentic desktop control. OpenAI’s own tool, which began as a developer-focused Codex system, was integrated into the ChatGPT app and introduced background computer use on the Mac earlier in the same year. That move was widely seen as a signal that the leading AI labs believe the next major battlefield is not chatbots but agents that can accomplish real work.
In this emerging competition, the design choices around computer control could be as important as the underlying model. Users have been clear that they want AI systems that assist without interrupting their flow. A feature that requires the user to stop everything and watch an AI move a mouse is not very compelling. One that quietly takes care of requests in parallel is far more attractive.
Anthropic’s upgrade positions Claude to be an effective partner in a busy working day. For Mac users subscribed to Pro or Max plans, this means a request such as “Find the documents that need signatures and prepare an email” can now be handled without turning the entire computer over to the AI.
Security and supervisory concerns
With deeper access and more independence comes greater responsibility. Allowing Claude to operate in the background increases the importance of safeguards. Users need to be confident that the AI will not alter files, send messages, or execute commands outside of the assigned task. Anthropic has emphasized safety measures for computer use since its inception, but the background mode may require additional protections.
For example, if Claude is working in a folder while the user is working in the same folder, there could be conflicts. A poorly written instruction might cause the AI to delete important files or modify documents that are open elsewhere. These risks mean that background computer use is likely to be offered first to users who are already paying for premium access and who have agreed to various terms of use. It also suggests that users should give Claude clear, well-defined tasks and periodically verify the results.
Anthropic has not yet revealed exactly how the model handles permission requests in the background. There are likely to be controls that allow users to review actions or interrupt the process if something goes wrong. The inclusion of such guardrails will be important as more organizations begin to rely on AI agents.
The road to agentic desktops
Today’s announcement is a reminder that computer use has evolved from a research demonstration to a product feature with real applications. In less than a year, the ability for an AI to look at a screen and operate applications has moved from an early beta to a practical option for subscribers on Mac. That pace is likely to continue as models become more reliable and as users help train the systems through real feedback.
For now, the arrival of background computer use in Claude Code and Claude Cowork gives Mac users a new way to think about delegation. Instead of having to watch Claude make progress in real time, they can simply let it run. That may be enough to convince more people that the age of AI agents is not something for the distant future, but a tool available today.
Source: 9to5Mac News