Claude for macOS Can Now Take Over Your Desktop and Run Tasks While You Are Away
If you use Claude Pro or Max on a Mac, this is a meaningful step change in what the AI can actually do for you. Anthropic has launched Claude Computer Use for macOS in research preview, allowing Claude to directly operate your desktop (opening apps, navigating your browser, filling in spreadsheets, and running developer tools) without you needing to stay at the keyboard. When Claude lacks a direct connector to an app, it falls back to pointing and clicking the screen, much as a human assistant would. The feature works in both Claude Cowork and Claude Code, and pairs with Dispatch (launched last week) which lets you queue tasks from your iPhone and return to find the work done at your desktop.
Safety prompts, halt commands, and prompt-injection scanning are built in, though Anthropic is clear this remains a research preview with limitations.
Claude Code Gets Auto Mode: the AI Now Decides on Its Own Which Actions Are Safe to Take
For developers using Claude Code, auto mode removes one of the more frustrating trade-offs in agentic AI: having to choose between micromanaging every action or turning the model fully loose. Anthropic's new auto mode, now in research preview, lets Claude review each proposed action before executing it, checking for risky behaviour the user did not request and scanning for prompt injection attempts. The practical effect is that Claude Code can run longer, more complex coding tasks without constant manual approval gates, while still maintaining a meaningful safety layer.
The feature currently works only with Claude Sonnet 4.6 and Opus 4.6, and Anthropic recommends using it inside isolated environments for now.
Gemini on Google TV Gets Richer Visuals, Sports Briefs, and Educational Deep Dives
Google has begun rolling out a significant visual upgrade to Gemini on Google TV across the US and Canada as of 24 March. If you use a Google TV device and ask Gemini about a sports match, you will now see live scorecards and broadcast options alongside the text response. Recipe queries pull in accompanying video tutorials. Educational Deep Dives (narrated visual breakdowns covering health, economics, technology, and similar topics) are also being introduced, offering a more interactive alternative to plain text answers.
Sports briefs cover the NBA, NCAA basketball, NHL, MLB, MLS, and NWSL. Broader device support for Deep Dives and sports briefs is expected this spring, suggesting this is a phased rollout rather than a completed feature.
Industry Themes
The most striking pattern today is Anthropic's move to give Claude physical agency over the computer. Both Claude Computer Use for Mac and Claude Code auto mode, announced the same day, represent a shift from Claude as a conversational assistant to Claude as something closer to an autonomous worker that operates software on your behalf. These announcements surfaced on Anthropic's own channels before mainstream press coverage, which is worth noting for anyone who tracks AI developments in real time.
The second theme is the battle for the living room, with Google pushing Gemini further into Google TV via richer visual responses and sports intelligence. As AI assistants move beyond smartphones and browsers into ambient home devices, the interface patterns (visuals, narration, real-time data) start to look quite different from a chat window, and Google appears to be iterating on what that looks like in practice.
Finally, the quiet from OpenAI, xAI, Meta, and Mistral over the past 24 hours is itself notable. After a dense first half of March for model releases, the field appears to be in a brief consolidation phase, though several major announcements (Grok 5, GPT-5.5 speculation, Mistral's next flagship) remain expected in the coming weeks.