Voice, mobile and desktop: same Claude, different affordances

Which mode answers "chat, project, or agent?" This page answers the other axis: web, desktop app, phone, or voice — where the model is the same and everything around it differs. The differences that matter are what a surface lets you put in, and what it lets you check.

Voice: hands free, eyes free, review deferred

Voice is a real mode now, not a dictation shortcut, and three of its properties are worth knowing before trusting it with anything load-bearing.

It is the same model and the same meter. Voice "starts with the model you last used in text chat", and "Voice mode can use the same Claude models you use in text chat, so it can keep up with harder questions, not just quick ones". It also draws from the same budget: "Voice conversations count toward your regular usage limits based on your subscription plan" — pace yourself accordingly on a capped plan.

Your connectors come with you. "Connected tools work the same way in voice mode as they do in text chat and follow the rules of your plan". That sentence cuts both ways: the convenience of "check my calendar" while cooking, and every grant question now answerable from your pocket, in a mode where you are not looking at what happens.

Review is the weak link, by design. The docs are candid that "Not every result can be shown on screen in voice mode, and using several tools at once can add a short delay". A spoken answer cannot be skimmed, re-read, or diffed — which makes voice the right surface for asking and thinking, and the wrong one for anything whose answer you would normally verify line by line. Hear it, then read it later, is a fine workflow; act on a number you heard once is not.

Mobile: full input, thin verification

The phone app is complete as a chat surface — camera and files in, voice in and out, connectors live. The structural limit is on your side of the screen: a phone is where fluent answers get skimmed, and skimming is where confident errors pass. The habit that compensates costs nothing: on mobile, collect and ask; defer decide and use to a screen where you can hold the answer still.

Desktop and web: where checking is cheap

The web app is the baseline. The desktop app's difference is reach into the machine: for agent work, "Desktop is the full Cowork experience, where Claude can also use your local files and browser" — capability and blast radius arriving together, as always. Desktop is also where side-by-side reading, long documents and project work are ergonomically honest: the surface you verify on.

The corollary is a simple placement rule: gather anywhere, verify on glass, grant on desktop. Questions and capture fit every surface; reviewing output needs a real screen; and decisions about connectors and file access deserve the surface where you can see their consequences.

What goes wrong

Acting on the heard number. The figure was misheard, or spoken from a confidently wrong answer, and no record of the difference exists until the consequence does.

Voice as dictation for precision work. A contract clause or a config value through speech-to-text inherits transcription noise on top of model noise, invisibly.

Tool sprees from the pocket. Voice plus connectors means real actions in your accounts from a surface built for glances. Keep the grants narrow on the account, and the surface stops mattering.

The phone-approved diff. Anything that would get a careful read at a desk gets a thumb-scroll on the train, and ships.

Surface-hopping mid-task. Half the context in a voice chat, half in a desktop conversation, and neither knows the other happened. Conversations do not follow you across unless you carry them.

How to check it worked

Audit one week of your own usage by decision weight: list the moments a Claude answer changed something real (money moved, message sent, work submitted) and note which surface each ran on. The pattern to want: consequential moments cluster on surfaces where you could and did verify. If the heavyweight decisions are happening by voice and thumb, the fix is not less mobile use — it is moving the last step of those tasks to glass.