As of November 12, 2026, Google's asynchronous coding agent Jules has officially entered read-only mode ahead of its permanent shutdown on December 31.

According to the official turndown notice in the changelog, Google is consolidating its agentic coding efforts into Google Antigravity.

I use Antigravity every day, and as an interactive local agent environment, it is great. But telling Jules users to "just use Antigravity" glosses over the one architectural feature that made Jules genuinely special: every task ran inside its own isolated cloud virtual machine, completely decoupled from your local hardware.

Most AI coding tools—Cursor, Claude Code, Windsurf, and Antigravity—treat agentic coding as a local, tethered activity. Even when an agent runs background terminal tasks or spawns subagents, your local machine is still the host. Your CPU is spinning, your battery is draining, your network connection has to stay alive, and your laptop has to stay awake.

Jules worked on a completely different mental model:

  1. Write down a list of well-scoped engineering tasks, refactors, or bug fixes.
  2. Dispatch all of them at once straight from the CLI.
  3. Close your laptop lid and go do something else.

While you were at the gym, grabbing coffee, or sleeping, Jules spun up dedicated cloud VMs in the background, cloned your GitHub repositories, installed dependencies in clean sandboxes, ran your test suites, and opened pull requests. When you opened your laptop again, the work was sitting on GitHub waiting for review and merge.

You weren't babysitting a terminal window or worrying that closing your MacBook or Windows laptop would kill three half-finished agent runs.

Consolidating products makes sense from Google's perspective, and plenty of Jules's planning and MCP capabilities now live inside Antigravity.

Still, there is currently no native option in Antigravity to offload heavy batch work off your laptop. If you want five parallel tasks running across different branches, your local machine has to carry the compute, disk I/O, and active session state for all five. You can't just fire a CLI command from a lightweight terminal, shut your computer down, and let cloud VMs chew through your backlog.

Losing a tool that genuinely freed you from your desk stings a bit. That said, agent workflows are evolving fast right now—between headless CLI agents, ephemeral cloud sandboxes, and MCP-driven pipelines, I am looking forward to seeing what new ways we find to work with autonomous agents next.