If you spend your days living inside a code editor, Anthropic just gave you a reason to stop alt‑tabbing quite so much. Claude Code on desktop can now preview your running apps, review your changes before you push, and quietly babysit your CI checks and pull requests in the background – all from the same window. It’s a small quality‑of‑life update on paper, but it hints at where AI coding tools are really headed: from autocomplete sidekicks to full‑stack workflow copilots.

Let’s start with the most instantly satisfying bit: seeing your app run right next to the AI that’s helping you build it. With the latest update, Claude Code can spin up dev servers and embed a live preview of your web app directly inside the desktop client. It doesn’t just show a static iframe; Claude can actually “look” at the UI, read console logs, spot runtime errors, and iterate on the code without you having to narrate what’s broken. You can even click on a visual element in the preview and pass that context back to Claude, so instead of typing “the blue button in the top‑right is misaligned,” you just select it and say “fix this layout.” The promise is pretty clear: fewer context descriptions, more “just make it look like this” interactions.

Once the UI looks right, most teams hit the same next step: “okay, let’s clean this up before it leaves my machine.” That’s where the new Review code button comes in. Hit it, and Claude will read your local diffs, then layer comments directly into a diff view inside the app, pointing out obvious bugs, questionable patterns, and things that might bite you in production. It’s essentially an always‑available, never‑tired reviewer who doesn’t mind nitpicking your error handling for the tenth time this week. If you like what it suggests, you can ask Claude to apply the fixes, turning what used to be a back‑and‑forth with a teammate into a tight local feedback loop.

The more interesting shift, though, happens after you finally open a pull request. Instead of bouncing over to GitHub every few minutes to see whether CI is still red, Claude Code now tracks PR status for GitHub projects from inside the desktop app, using the GitHub CLI behind the scenes. You get a CI status bar inside your session, and when checks fail, Claude can read the CI logs, attempt a fix, and push updates automatically if you’ve enabled the auto‑fix option. Flip on auto‑merge as well, and once everything turns green, the PR is squashed and merged for you – assuming your repo allows auto‑merge in the first place. The net effect is that you can move on to a different task knowing there’s an AI janitor cleaning up after your tests in the background.

There’s a broader trend here beyond just one app. Other tools have experimented with similar ideas – OpenAI’s Codex CLI, for example, has a reference setup where failed CI runs automatically trigger an AI job that proposes a fix and opens a pull request with minimal, test‑oriented changes. What Anthropic is doing with Claude Code is folding that kind of automation directly into the everyday editor‑adjacent experience: you write code, you ask for a review, you open a PR, and the same assistant sticks around to shepherd it through CI instead of disappearing after the last keystroke. For teams already using GitHub, that’s a compelling reduction in tooling friction.

The other piece of the story is portability. Development workflows are increasingly multi‑device and multi‑surface, and Anthropic is trying to make Claude Code feel less like a single app and more like a session you can carry around. Start something in the CLI, run a command to pull that context into the desktop app, then, if you need to leave your desk, move the session to the cloud and pick it up later in a browser or on the Claude mobile app. Docs around Claude Code on the web highlight this “teleport” style of working: send a long‑running task to the cloud, let it continue on their servers, and reconnect from wherever you happen to be. It’s the same conversation, the same background jobs, just hopping between devices.

For developers, the pitch is straightforward: less toil, more flow. Kick off a feature in the afternoon, let Claude handle the boring bits – wiring up boilerplate, watching CI, nudging PRs over the finish line – and by the time you’re ready to context‑switch back, your branch might already be green and merged. Some in the ecosystem are already speculating that, at least inside AI‑heavy companies, the traditional “software engineer” role could morph into something more like a curator or orchestrator of AI‑written code, as assistants like Claude Code shoulder more of the typing and plumbing work.

Of course, this doesn’t magically solve every hard problem in software development, and nobody sane is arguing you should blindly trust auto‑fixed CI or auto‑merged PRs in production without guardrails. The safest setups will combine Claude’s automation with strict repository policies, test coverage, and human review on mission‑critical paths. But as these tools get better at reading logs, understanding context, and respecting constraints, the line between “assistant that helps me write code” and “agent that quietly manages my development lifecycle” is going to keep blurring.

For now, Claude Code’s new desktop tricks – live app previews, inline AI reviews, CI babysitting, and portable sessions – feel like another step toward that future. If you already live in an AI‑augmented workflow, this update is less about flashy new capabilities and more about stitching them together so you can stay in flow longer, context‑switch less, and let the background noise of modern software development fade into, well, the background.