Anthropic's Claude Cowork is an AI tool that sits inside the Claude desktop app and handles multi-step tasks on your computer: pulling data from emails and documents, organizing files, writing reports and drafting emails. The important difference from a normal chatbot is that you give it a goal and it keeps working without you having to guide every step.
Until now it had one major constraint. Your computer had to be on and the app had to be open for it to do anything. Some users simply left their laptops half-open overnight to keep sessions alive. That workaround is now gone. Cowork can continue working after you step away, which also means you can assign tasks before bed and come back to finished outputs in the morning.
Managing it from your phone is available via the Claude mobile app on both Android and iOS. To be precise about what this means: your phone does not run the tasks. Your desktop still does. The phone gives you a way to assign work, check status, and get notified when something is done. Think of it as a remote control for work happening back at your desk.
Anthropics own usage data tells a story worth paying attention to. Most Cowork usage comes from operations, marketing, finance, and legal teams, not from technical staff. The top task categories are business process and operations work such as data reports and checklists, and content creation such as slide decks and proposals. That pattern suggests these tools are moving into mainstream office work faster than most organizations have planned for.
This is not an isolated product decision. OpenAI launched a similar remote-control feature for its Codex tool in May, letting users manage desktop tasks from the ChatGPT mobile app. Google has its own always-on agent in progress. The shared direction is the same: AI that runs continuously in the background, managed from your phone the way you might manage a junior colleague via text message.
There is a real security question worth understanding here. When an AI agent runs in the background with access to your files, emails, and connected apps, it can also be tricked by malicious content it encounters. A hidden instruction embedded in an email or a document the agent reads can redirect what it does next. Security researchers classify this as the top vulnerability for AI agents in production environments. An agent with broad access to your systems does not just expose your own data if something goes wrong: it exposes the authority of every connected system it has permission to touch. This is not a reason to avoid these tools, but it is a reason to be deliberate about which folders, inboxes, and apps you connect.
For individuals, the practical question is whether the overnight-run and phone-management features make the tool more useful for your actual work. If you regularly hand off repetitive document tasks, data pulls, or report drafts, the answer is probably yes. Cowork is available on paid Claude plans starting at USD 20 per month, with the new mobile and background features rolling out to the higher USD 100 per month tier first.
For organizations, the more pressing question is governance. Anthropic has added admin controls for enterprise plans, including the ability to set which departments can access Cowork and spending caps by team. If your staff are already using this tool individually, and the usage data suggests many are, the decision has effectively already been made without a formal process. Getting ahead of that by setting access policies and data permissions is more practical than waiting.