For a while now, AI assistants have been very good at telling you how to do things. Today Anthropic decided that was taking too long.

Claude in Cowork and Claude Code can now take control of your computer to complete tasks. It opens your apps, navigates your browser, fills in spreadsheets, runs dev tools, clicks around your screen. Anything you would normally sit down and do yourself. You describe what you want, walk away, and come back to finished work. Your computer is now someone else's problem.

The way it works is clever. Claude reaches for the most precise tool first, starting with connectors to services like Slack or Google Calendar. When there is no connector available, it goes straight to your screen, controlling your browser, mouse, and keyboard directly. It takes screenshots to see what is on your screen, asks your permission before accessing apps, and warns you upfront that some actions cannot be undone. Which is a sentence worth reading twice before you hand it access to your files.

It pairs with Dispatch on mobile. You can assign Claude a task from your phone, put your phone down, and come back to the finished work on your computer. Tell it once to scan your email every morning or pull a report every Friday and it handles it from there, on a schedule, without being asked again. Essentially a personal assistant that works nights and weekends and has never once asked for a raise.

Anthropic is calling this a research preview and is upfront that it will not always get it right on the first try. Complex tasks sometimes need a second attempt, and working through your screen is slower than a direct integration. They are shipping early to get real feedback, which is either reassuring transparency or a sign that your spreadsheets are about to have an interesting week.

Computer use is available now for Claude Pro and Max subscribers on macOS. Update your desktop app and pair it with your phone to try it.

Your computer now has a coworker. It just never takes a lunch break.

Computer use is the headline today but if you zoom out for a second, something bigger is taking shape.

Buried in Anthropic's code, a developer spotted references to "Phone Use" alongside a feature called Orbit. The implication is that Claude is being built to operate your phone the same way it now operates your desktop. Make calls, send messages, execute tasks on your mobile device. Not just assist you. Act for you.

Look at what Anthropic has shipped or is actively building right now. Computer use. Browser control. Calendar management. Maps. File creation. Code execution. Phone calls. A task list system. Scheduled recurring actions. Memory across sessions. And now an agent layer called Orbit sitting in beta in the settings panel.

That is not a chatbot with extra features. That is an operating system for your life.

The comparison that keeps coming up is OpenClaw, the open source framework for building AI agents that can take real world actions. But what Anthropic appears to be building goes further. OpenClaw gives developers the tools to build agents. What Claude is being built to do works across your entire digital existence out of the box, desktop, phone, apps, calendar, files, the lot. No setup. No code. Just Claude.

None of this is fully live yet. Phone use is unconfirmed beyond a code reference. Orbit is still in beta. But the direction is unmistakable. Every capability Anthropic is adding points toward the same thing: an AI that does not wait to be asked, knows your context, and handles the work while you get on with your life.

A year ago Claude was a very good chatbot. Today it can use your computer. This time next year it might just run your entire day.

Jensen Huang appeared on the Lex Fridman podcast and said something that sent the internet into a spiral: "I think we've achieved AGI."

Before the existential panic sets in, it is worth understanding what he actually meant. AGI, or Artificial General Intelligence, has traditionally meant an AI that can do anything a human can do across any domain. Huang did not claim that. What he said is that if you define AGI as an AI capable of completing specific economic tasks, like generating a billion dollars in revenue autonomously, then we are already there.

That is a very convenient definition for the CEO of the company selling the chips that power every major AI system in the world.

Something real has shifted, and Huang is not entirely wrong that the capabilities today are remarkable. But declaring AGI by redefining what AGI means is not quite the same as achieving it. The field has never agreed on a single definition, which means anyone can move the goalposts to wherever the ball already landed.

The chips are selling either way.