Google Antigravity 2.0 lets you control AI coding agents from anywhere

Coding with AI agents has introduced a bizarre new workplace habit: developers sitting at their desks, anxiously watching terminal logs scroll by.

As agents take on massive tasks like full codebase refactors, end-to-end test suites, and complex version control, jobs can easily take half an hour or more. Stepping away means risking an agent sitting idle waiting for approval, but staying put feels like digital babysitting.

To solve this, Google rolled out Antigravity 2.0, headlined by a feature called Remote Control that untethers developers from their workstations.

Your entire setup inside any web browser

Remote Control gives developers a secure window into their running Antigravity sessions from any machine with an internet browser.

Instead of forcing engineers to mirror environments across multiple devices, the feature keeps the agent running on the original machine. That means all local build tools, project files, credentials, and environment variables stay exactly where they are, eliminating the pain of duplicate configs.

It also supports multi-instance management. A developer can keep a desktop dedicated to UI development, offload heavy backend tasks to a cloud server, and seamlessly switch between both sessions from a single laptop while working from a coffee shop.

To make untethered work truly practical, Google added proactive push notifications. Whenever an agent completes its turn or requires human input to resolve an issue, it pings the user immediately so tasks never get stuck in limbo.

Why it matters

We think Google took a remarkably pragmatic route here. Instead of forcing everyone to migrate their private code and complex setups into expensive fully-hosted cloud sandboxes, Remote Control turns your existing hardware into an on-demand engine that you can pilot from anywhere.

For senior developers and solo builders, this transforms AI from an interactive autocomplete tool into an asynchronous digital colleague. You fire off a massive refactor, walk away to grab lunch, and approve checkpoints from your phone or browser when alerted.

This shift signals a broader trend in AI software engineering: raw code generation is commoditizing, and workflow orchestration is becoming the real battleground. Rivals will need to offer similar asynchronous, multi-device handoffs to keep up.

That said, opening up direct web-based tunnels to developer machines carrying production credentials creates an obvious attack surface. The convenience is huge, but teams will want to inspect Google's authentication layers closely before turning workstations into remote-accessible nodes.