Vibe Check: Claude Code Now Works on Mobile and the Web

Claude Code has been transformative for us at Every, so I’m teaching a course for on getting started and using it for beginners—no coding experience required. Join me on November 19 for a live, full-day workshop where you’ll install Claude Code on your machine and build an app end to end. Early bird pricing is $1,000 (and includes an annual Every subscription) until October 23. Learn more and save your seat.—Dan Shipper

Was this newsletter forwarded to you? Sign up to get it in your inbox.

I'm in San Francisco for back-to-back conferences. Kieran Klaassen is chasing toddlers and going on car trips upstate. Danny Aziz is in the middle of a cycling expedition. This is when we need AI coding tools that work from anywhere—when life is happening and the laptop isn't always within reach.

I’ll admit that I've been using OpenAI's Codex for most of my coding lately, but Kieran still swears by Claude Code, running multiple instances in parallel as part of his compounding engineering workflow. So when Anthropic gave us early access to the cloud and mobile version of Claude Code that just launched, we all carved out time to test the same promise: Can you actually run Claude Code from your phone? Can you assign it a coding task while you're away from your desk and pick it up later? Does "anywhere" truly mean anywhere?

Coding manually from your phone has been possible in theory, but tiny screens, clumsy keyboards, and not having anywhere to run and test your code mean that I don’t really use it much. Danny (general manager of Spiral) has his own Claude Code setup running on a server that seems to work for him, and Monologue general manager Naveen Naidu uses Codex on the web to handle small papercut bug fixes.

But the dream is being able to kick off a Claude Code session on my laptop and then keep it running in the cloud when I go to a meeting. So my question was: Is Anthropic's cloud and mobile version the one that finally delivers?

Short answer: Not yet.

We spent the weekend testing across three different real-world scenarios, and while the underlying idea is sound, it’s not something we’d use every day yet. It feels very much like Codex on the web: You can begin a task on web or mobile, it runs your code in a virtual machine, and you can keep chatting with it as it does its work. But getting the cloud environment to handle our setup was a pain, the mobile app lacks important features like inline code diffs, and most importantly, you can’t kick off a task on your computer and then transition it to the cloud.

A caveat: This version is a work in progress. Anthropic is leaning into AI's new normal—ship early, iterate fast, improve in public. What we tested will likely be better by the time you read this, and better still in a few weeks. When we shared our feedback with the team (as we do when we’re given these advance previews), they responded immediately and are actively fixing many of the issues we hit like our problems with the cloud environment. So while in its current state, we won’t reach for it every day, we feel excited about where it could be in a few weeks or months.

So what’s new?

The cloud and mobile version of Claude Code is Anthropic's answer to remote coding. Like OpenAI’s coding agent Codex, it promises to spin up virtual machines in the cloud with your codebase, letting you delegate coding tasks without keeping your laptop open.

Here's how it works. There’s a new tab in the Claude app labeled Code:

When you click it, you’re given a text field where you can describe what you want Claude to do:

It boots up a cloud environment with your repository, sets environment variables (like which domains it can access), and starts working. You can watch its progress, see its to-do list, and—here's where it differs from Codex—keep chatting with it while it works. In Codex, you have to stop execution to talk to the agent. Claude Code's interface treats it as an ongoing conversation.

Normally, when you run Claude Code on your laptop, the AI needs your computer to stay on and connected; it's using your machine's processing power and your local copy of the code. "Cloud" means Anthropic runs everything on its remote servers instead. Your code lives there, the AI runs there, and the work happens whether your laptop is open or not. You can check progress from your phone, pick up where you left off from a different computer, or just walk away while Claude keeps working.

Existing AI-powered coding tools like Codex and Replit already promise cloud execution and mobile access. But Codex can't hold a conversation while your code runs, and Replit's mobile app is plagued with blank screens and keyboard failures. If Claude Code Cloud could nail both—let you chat with it mid-execution and deliver a mobile experience that works—it would solve the two biggest pain points in the category.

For this vision to deliver on its promise, it needs a few core ingredients:

1. True mobile access. The app needs to let you browse code repositories, initiate tasks, and check on your progress from your phone. Otherwise, you're back to being chained to your laptop.
2. Workflow portability. Your custom Claude Code slash commands (quick shortcuts like /review or /work that trigger specific actions) and agents (specialized AI workers with designated roles) should work identically in the cloud version, so your carefully crafted local workflows translate. Otherwise, where’s the efficiency gain?
3. Seamless handoffs. You should be able to start a task on your laptop, close it, and let Claude keep working in the cloud while you live your life, then check progress from your phone later. Without that, you're still tethered to one device at a time, defeating the whole "work from anywhere" premise.

That’s the bar. Now let’s look at how well it meets it...

Become a paid subscriber to Every to unlock this piece and learn about:

1. Great concept, buggy execution
2. Environment setup failures block productivity
3. Mobile-first vision undermined by validation limitations
4. Conversational strength can't overcome functional weaknesses
5. The results of the Reach Test from Dan, Kieran, and Danny