Malhar Ujawane

Inspiration

Inspiration·

Google is shutting down Jules on December 31

Jules goes read-only after Nov 11, 2026 and shuts down Dec 31. Google's export runs through a CLI agent like Claude Code. Keep agent context in the repo.

Google is winding down Jules, its asynchronous coding agent. You can keep working in sessions until November 11, 2026, after which Jules turns read-only, and it shuts down on December 31. Google says the learnings are going into Google Antigravity.

The changelog is clear about what you keep:

Your code, branches, and pull requests already live in GitHub, so they stay where they are. The export covers the configuration and history that live inside Jules.

That export covers account settings, environment variables, scheduled tasks, connected repos, memories, setup scripts and every session’s prompt, plan, activity log and patches. The interesting part is how you run it. Google tells you to install @google/jules, log in, then “open a CLI coding agent, such as Claude Code” and paste a long export prompt. That prompt has the agent read your Jules token out of the macOS Keychain or secret-tool, call endpoints the prompt itself calls undocumented, and write everything to ./jules-export. It even pre-empts the agent’s habits:

Both are undocumented, so if a call starts returning 404/400, tell me which one rather than guessing at fixes.

Two things I take from it. The first is that Google’s own way out of Jules runs through a coding agent, and the example it names is a competitor’s. The second is that it puts an agent next to an access token and a file of environment variable values. The prompt tells the agent not to print either, and Google tells you to keep the export folder out of your repositories. I’d read the script before pasting it, the same way I’d read any MCP server before connecting it.

The bigger lesson is about where an agent’s memory lives. Anything Jules learned about your repo, its setup scripts and test commands, sat in Google’s database, and now it’s a JSON file you have to go and fetch. If that context lives in the repo as an AGENTS.md, a setup script and a test command, switching agents costs nothing. Plain git is the most portable place for agent work, which is also why I like worktrees for parallel agents. Theo’s line about dashboards he won’t open is the user-side version of the same preference.

Get new posts in your inbox

© 2026 Malhar Ujawane. Built with Astro.