Anthropic Tests 'Orbit': Claude Is Coming to Your Phone
Anthropic is reportedly testing "Orbit," an experimental feature that would let Claude operate your smartphone the way Computer Use already operates your Mac and Windows PC, tapping, typing, and navigating apps on your behalf. Here is what was actually found, what is still unconfirmed, how it compares to Google and Apple's on-device agents, and what it would mean for the way you use your phone.
In this article

⚠️ Updated April 30, 2026: Claude's Computer Use feature launched on macOS (March 28, 2026) and expanded to Windows (April 14, 2026) — both in research preview. See Claude Can Now Click, Type, and Navigate Your Mac for the latest. Orbit (mobile phone control) has not yet been released publicly.
How Orbit Was Discovered
It is worth being precise about what is actually known here, because the gap between "a feature name found in code" and "a product Anthropic is shipping" is large. What surfaced was a reference to "Orbit" inside Anthropic's own codebase — the kind of internal name that engineers attach to a project long before it has a launch date, a pricing page, or even a public commitment. Companies routinely build, rename, and quietly shelve features that reach this stage. So treat Orbit as a credible signal of direction, not a confirmed roadmap item.
Why does the signal carry weight anyway? Because it lines up with a pattern Anthropic has already made public. The company shipped Computer Use — Claude reading the screen and controlling a desktop with a virtual mouse and keyboard — first on macOS, then on Windows, both as research previews. A mobile equivalent is the obvious next surface. Phones are where most people actually live their digital lives, and an assistant that can only act on a laptop leaves the most-used device untouched. Orbit, whatever it ultimately becomes, points squarely at closing that gap.
Why a Phone Is Harder Than a Desktop
On a Mac or PC, Computer Use leans on decades-old conventions: a visible cursor, resizable windows, a filesystem, and screenshots that an AI can interpret frame by frame. Mobile throws most of that away. There is no mouse pointer, apps run full-screen one at a time, and the operating system aggressively sandboxes them so one app cannot easily see or touch another. That sandboxing is a security feature for users, but it is exactly the wall an autonomous agent has to get over to be useful.
In practice, an agent like Orbit would most plausibly work the way assistive technology already does: by hooking into the platform's accessibility APIs, the same layer that screen readers use to identify on-screen elements and trigger taps. That is a well-trodden path, but it comes with hard limits set by Apple and Google. iOS in particular keeps a tight grip on what background processes can automate, which is why a fully autonomous, install-and-go phone agent may arrive on Android first — or require deeper OS-level cooperation than a third party like Anthropic can unilaterally build. This is the central technical reason Orbit could stay experimental far longer than its desktop sibling did.
How Orbit Would Compare to Google and Apple
Anthropic would not be entering empty territory. Google has been building Gemini into Android as a system-level assistant and has demonstrated agentic, multi-step task automation in the browser through its Project Mariner research. Apple, meanwhile, has pushed on-device automation through Apple Intelligence and a more capable, app-aware Siri. The strategic difference is positioning: Google and Apple control the operating system and can wire an agent directly into it, while Anthropic is a model company that would have to operate through whatever doors the platform owners leave open.
That is also Anthropic's potential edge. A device maker's assistant is loyal to that device's ecosystem; a model-layer agent can, in principle, drive any app on any platform with the same underlying intelligence. If Orbit ships, the pitch is not "a better Siri" but "the same Claude you trust on your desktop, now on your phone" — one assistant spanning browser, computer, and mobile rather than three siloed helpers.
What It Could Actually Do for You
- Triage and reply to texts or DMs in a messaging app it does not have an official API for
- Book a ride, order food, or check in for a flight by driving the app's own screens
- Reschedule a calendar conflict and notify the affected people in one chain of steps
- Pull a confirmation number out of one app and paste it into a form in another
- Run a repetitive multi-tap workflow — like logging a daily reading into a health app — on a schedule
🔒 The flip side: an agent that can tap anything on your phone can also tap the wrong thing. A mobile agent would need access to messages, contacts, banking, and payment apps to be useful — and that same reach is what makes a misfire or a prompt-injection attack costly. Expect Orbit, if it launches, to ship with explicit per-action confirmation and tight permission scoping rather than free rein. Anthropic's decision to keep desktop Computer Use in research preview suggests the same caution will apply here.
The Takeaway
Orbit is not a product you can use today — it is an internal reference that tells us where Anthropic's thinking is pointed. The direction is consistent and clear: Claude is being built to act, not just answer, across every screen you own. Computer Use proved the desktop case in research preview; the browser is already in play; the phone is the logical and most valuable next step. The open questions are not really about ambition but about constraints — how much autonomy mobile operating systems will permit a third-party agent, and how Anthropic balances usefulness against the risk of handing an AI the keys to your most personal device. Until there is a public release, treat Orbit as a window into Anthropic's roadmap, and watch whether the next Computer Use update mentions mobile at all. We will update this article the moment it does.
Get the Claude playbook in your inbox.
One weekly email for Claude and Claude Code users. Real workflows, no hype. Subscribe and we send you The Claude Power-User Cheatsheet.
— ¶ —

Luke Thompson
Luke Thompson is the founder of The Operations Guide, LLC and editor of The Claude Insider. Based in Jonesborough, Tennessee, he has spent years building AI-augmented business systems and automation workflows for operators and teams. He began working with large language models in production well before the current wave of consumer AI tools, integrating them into client workflows, content pipelines, and operational infrastructure. At The Claude Insider, he writes about Claude with the specificity of someone who uses it daily as a professional tool — not as a reviewer or commentator, but as a builder. His coverage focuses on what actually works: prompt patterns, API integration strategies, agentic workflows, and the real-world tradeoffs that practitioners face. He is not affiliated with Anthropic, PBC.
Articles are researched and drafted with AI assistance, reviewed and edited by Luke Thompson.
Know where AI can pay off in your company.
Take the free two-minute AI Readiness Assessment. See your score, the two gaps holding you back, and the next move worth making.


