Google Antigravity Remote Control: Your Coding Agent, Untethered

Agentic coding has quietly changed shape over the past year. Developers no longer prompt for a snippet at a time — they delegate entire subsystem refactors, large-scale test suite runs, build-breakage diagnoses, and version-control operations across whole workspaces. The consequence: agent tasks that run for tens of minutes or hours, while the human who launched them sits tethered to a desk, watching logs scroll. On August 21, 2026, Google’s Antigravity team addressed exactly that pain point with the launch of Remote Control for Antigravity 2.0 — and it may be the most practically useful agentic-dev feature shipped this month.

The problem: agents that outlast your attention span

The Antigravity team’s framing in the announcement is blunt. Traditionally, a developer working with an agent had two bad options: stay at the workstation staring at agent logs, hoping the task finishes sooner, or step away blind — with no way of knowing whether the agent finished five seconds after you left or is still grinding twenty minutes later.

That’s not a rounding error in developer experience; it’s the core bottleneck of supervising long-running autonomous work. The entire value proposition of coding agents is that they work while you don’t. If you must physically babysit the machine, the productivity gain collapses. Remote Control is Google’s answer: carry your agents with you as you move, instead of being tied to a single physical workstation.

What Remote Control actually does

Remote Control lets you securely connect to and drive your Antigravity sessions running across your different machines — laptops, servers, desktops — from nothing more than a web browser. The docs describe it as “a secure window into your workspace,” and the emphasis is deliberate: this is not a cloud copy of your session, but a second client connecting into the session already running on your host machine.

Google highlights four benefits:

1. Multi-instance management. Dedicate specific machines to specific work — UI development on a local laptop, server-side tasks on a cloud environment — and switch between them seamlessly from a single browser dashboard, managing multiple environments and agents at the same time.
2. Untethered productivity. Start a complex task and walk away. From anywhere, you can monitor progress, review changes, and execute commands on the host machine.
3. Local context retained. No need to recreate your build environment on whatever device you happen to be carrying. Your files, workspaces, build tools, credentials, and environment variables stay on the host; the remote client simply reaches into them.
4. Proactive push notifications. The killer detail. If you’ve installed the web app to your phone’s home screen, Remote Control fires a push notification the moment an agent finishes a turn and needs your input. The agent effectively taps you on the shoulder from across the city.

How it works under the hood

Enabling the feature is straightforward. In Antigravity 2.0, open Settings (Cmd+, / Ctrl+,), navigate to Account, and toggle “Enable Remote Control” — optionally setting a nickname like workstation-primary so the machine is recognizable in your instance list. Then, from any browser, sign in to the Remote Control dashboard with the same Google Account, pick the machine from the instance switcher, and you have full access to active conversations, new agent tasks, implementation plans, and artifacts.

For headless machines there’s a daemon: a one-line installer (curl -fsSL https://antigravity.google/cli/agy-daemon.sh | bash on Linux/macOS, with a Command Prompt variant for Windows) registers a service that starts at boot, survives sign-outs on Linux, and re-authenticates itself after reboots following a single one-time terminal sign-in. The docs are unusually candid about operational nuances — for instance, the daemon has its own sign-in separate from the editor’s, and Windows uninstalls require an Administrator prompt.

Two architectural constraints are worth noting. First, the host must stay awake and online; if your network drops briefly, the web client reconnects automatically and background tasks keep running on the host uninterrupted. Second, it’s browser-based — there is no dedicated Google-made iOS or Android Antigravity app. On mobile you install the web app to your home screen for push notifications. That’s a philosophical choice: any device with a browser works, at the cost of native-app polish.

The competitive context: a two-horse race for remote agent control

The timing is what makes this launch notable. One day before Google’s announcement, Anthropic shipped a reliability overhaul of Claude Code’s mobile remote control, adding device cards and a local-only connection mode. Within roughly 24 hours, the two most prominent agent harnesses in the industry landed features aimed at the same problem — controlling your coding agent from somewhere else — but with opposite architectures.

Antigravity Remote Control is browser-first and explicitly multi-machine: one dashboard, many hosts, switching between instances. Claude Code’s remote control lives in a native mobile app and, as of its August 21–22 update, defaults to a local-only connection with no cloud relay — arguably a stronger security posture for the traffic path, at the cost of flexibility. Community reaction to Google’s launch has characterized it as “catching up” rather than pioneering — third-party remote-control hacks for Antigravity have circulated on GitHub since early 2026 — but catching up with an official, supported, multi-instance implementation is precisely what mainstream adoption requires.

Why it matters

Rollout began August 21 to all users, with Google AI Ultra subscribers getting priority access, bundled with Antigravity 2.0. The deeper significance is what it signals about the trajectory of developer tooling: as agents take on longer-horizon work, the locus of supervision is shifting from the IDE to whatever screen is nearest to the human. Push notifications for “agent needs input” are, functionally, the beginning of a world where your coding workforce pings you like a colleague on Slack.

The desk was the last place a developer’s agent had to stay. As of this week, it doesn’t.