Softwr

News

Google is shutting down Jules: read-only from 11 November, gone on 31 December

Google announced on 29 October 2026 that Jules, its asynchronous AI coding agent, is closing. Users have until 11 November to finish active sessions, Jules is read-only from then until the end of the year, and it winds down on 31 December. Code, branches and pull requests are safe in GitHub. Everything else Jules held, from scheduled tasks and memories to environment variables and session history, has to be exported with the Jules Tools CLI, and anything in your pipeline that calls Jules needs replacing within thirteen days of the announcement.

By Softwr Editorial, Software research teamPublished 11 min read
Share

The short answer

  • Google announced on 29 October 2026 that Jules will wind down on 31 December 2026, and users have until 11 November 2026 to finish active sessions before the service turns read-only.
  • Code, branches and pull requests already live in GitHub and are unaffected; account settings, environment variables, scheduled tasks, connected repositories, memories, setup scripts and every session must be exported with the Jules Tools CLI before 31 December.
  • The 11 November date is the one that matters operationally: anything that creates Jules work, including scheduled tasks, CI jobs that call the Jules API and pull request automations, should be treated as stopping that day, 13 days after the announcement.
  • Google says the work is folding into Antigravity, but as of 30 October it has not said when Jules capabilities will be available there, and the Antigravity product page describes local agents rather than the hosted asynchronous sessions Jules ran.
  • The shutdown notice says nothing about refunds or about changes to the Google AI Pro and Ultra plans that raised Jules task limits, so subscribers should not assume either will happen automatically.

What Google announced, and the three dates

Google posted the news as an entry headed "Jules Turndown" in the Jules changelog on 29 October 2026. It is short, and the key sentence is unambiguous: "Jules will wind down on December 31, 2026." The entry describes Jules as an experiment in how asynchronous coding agents could help developers work through a backlog, and credits it with shaping Google's other agentic products, naming "our agent-first platform, Google Antigravity."

There are three dates, and they do different things.

DateWhat happensWhat it means for you
29 October 2026Shutdown announced in the Jules changelogPlanning starts now. There is no earlier warning period.
Until 11 November 2026Users can continue finishing up sessionsLast day to complete work in progress. Treat it as the last day Jules does anything new.
12 November to 31 December 2026Jules is read-only until the end of the yearHistory is visible and exportable. Scheduled tasks, API calls and new sessions should be assumed not to run.
31 December 2026Jules winds downAnything not exported is gone. Export well before this date.

Thirteen days separate the announcement from read-only. That is short for anything wired into a build pipeline, and it is the figure buyers should plan around rather than the year-end date most headlines lead with.

What the notice leaves out

The changelog entry does not mention refunds. It does not mention the Google AI Pro or Ultra plans, which the same changelog said in August 2025 were the route to higher Jules task limits. It does not separately address the Jules API, which Google launched in October 2025 and explicitly suggested wiring into CI/CD pipelines such as GitHub Actions. And according to Times of AI's reporting the day after, Google has not announced a date for when Jules capabilities will be available through Antigravity, nor confirmed whether session histories, memories or scheduled tasks will transfer.

Those are questions a paying customer should put to Google in writing rather than assume an answer to.

The signal that came before

The changelog itself carried a hint. Through 2025 and early 2026 it recorded a steady run of features: memories in September 2025, the Jules Tools CLI and the API in October 2025, scheduled and suggested tasks in December, MCP support in February 2026 and automatic fixing of failed CI checks later that month. The last feature entry before the turndown is dated 9 March 2026, adding Gemini 3.1 Pro. Nearly eight months of silence on a product's own changelog is worth noticing, and it is the same pattern we saw across the closures in our 2026 shutdown list.

What survives, what you must export, and how

The good news is that the part of Jules most people care about is already safe. Google's notice is direct: "Your code, branches, and pull requests already live in GitHub, so they stay where they are." Jules worked by opening branches and pull requests against connected GitHub repositories, so merged work, open pull requests and their review history are untouched by the shutdown.

What lives only inside Jules is the configuration and the history. Google lists the export as covering:

  • account settings
  • environment variables
  • scheduled tasks
  • connected repositories
  • memories
  • setup scripts
  • every session, including its prompt, plan, activity log and code patches

How the export works

The export runs through Jules Tools, the command line client Google released in October 2025. The steps Google gives are to install it with npm install -g @google/jules, run jules login, then open a command line coding agent in your home directory (the page suggests the Antigravity CLI) and paste an export prompt published on the changelog page. The agent checks the setup, runs the export and reports what it saved.

Two cautions are worth stating plainly.

First, the export includes a file of environment variable values, and Google warns that it contains secrets and should be kept private and "out of your repositories." In practice that file is a list of credentials an agent will write to disk on your machine. Store it somewhere encrypted, and rotate any key that you are not certain has stayed private.

Second, the export is a prompt handed to a coding agent that will run commands and read your Jules login. Read the prompt before you paste it, and check what it wrote afterwards.

What the exported data is good for

Be realistic about this. An export of memories, setup scripts and environment variables is useful raw material for configuring whatever replaces Jules: setup scripts translate fairly directly into the environment configuration every alternative below needs, and memories are effectively notes on how your repository should be worked on. None of it is an import file for another product, and Google has not said that Antigravity will accept it.

The checklist before 11 November

For a single developer who used Jules from the web interface, the job is an export and a choice of replacement. For a team that wired Jules into its workflow, there are four places it can be hiding.

1. Finish or abandon open sessions

Any session still running needs to reach a pull request by 11 November. Review what is in flight now. A session that produced a plan but no code is cheaper to restart elsewhere than to rush.

2. Inventory scheduled tasks

Scheduled tasks arrived in December 2025 and are the easiest thing to forget, because they run without anyone opening Jules. Export them, then write down what each one did: dependency updates, test generation and recurring clean-up are the typical jobs. Each one needs an owner and a new home, or a deliberate decision that it is no longer needed.

3. Find CI jobs and scripts that call Jules

The Jules API launched on 3 October 2025, and Google's own announcement suggested integrating it into pipelines "like GitHub Actions." Search your repositories and GitHub Actions workflows for the Jules API, for the @google/jules package and for any stored Jules API keys. Remove the jobs or switch them off before 11 November so that they fail visibly in a branch you control rather than silently in a scheduled run, then revoke the keys.

4. Check pull request automation and branch rules

From February 2026 Jules could pick up failing CI checks on its own pull requests and push fixes. Teams that relied on that loop should expect open Jules pull requests to stop being repaired. Look for branch protection exceptions, bot accounts or commit authorship settings added for Jules, and remove them once the last Jules pull request is merged or closed.

5. Raise the billing question

If you pay for Google AI Pro or Ultra partly or wholly for Jules, the notice gives no indication of a refund or a plan change. Those plans include other Google AI products, so a refund is unlikely to be automatic. Ask, keep the answer in writing, and make the renewal decision on what the plan still includes after 31 December.

Antigravity, Google's answer, and what it does not yet answer

Google's stated destination for this work is Antigravity, which its own site describes as "our agentic development platform." It is a family of products rather than a single tool: a desktop app for managing multiple local agents in parallel, a CLI for running autonomous coding agents and background subagents from the terminal, an agentic IDE and a Python SDK. For individual developers the site says it is "Available at no charge," on macOS, Windows and Linux.

For a Jules user the important word on that page is "local." The Antigravity app is described as a command centre for local agents. The page does not describe hosted, remote execution, does not mention GitHub integration and does not mention Jules. Jules was valuable precisely because it ran in a Google-hosted virtual machine against a GitHub repository while your laptop was closed, and that is the part of the workflow for which Google has not published a replacement.

Google may yet announce a hosted mode or a migration path, but as of 30 October there is no date for either. Judge Antigravity as it currently exists, not on what it may become.

Who should still pick Antigravity

A developer who mostly used Jules interactively, starting a task and checking back within the hour, loses little by moving to an agent that runs locally, and Antigravity costs nothing to try. For teams that need agents running unattended against a shared repository, from CI or on a schedule, the hosted options below are closer to what Jules did.

The alternatives, with honest trade-offs

Five products cover most of what Jules users relied on. They differ less in whether they can write code, which all of them can, than in where the agent runs, which source control it supports and how the usage is billed. The figures and limits below come from each vendor's own documentation as of 30 October 2026.

ToolWhere the agent runsSource controlThe catch
Google AntigravityYour machine (local agents)Not stated on the product pageNo published hosted mode or Jules migration date
OpenAI CodexCloud workspace per task; keeps running while your computer is offGitHub or GitLabTied to a ChatGPT plan; environments need setting up
Claude CodeTerminal, IDE, or cloud sessions on Anthropic-hosted virtual machinesGitHub for cloud sessionsNeeds a paid Claude plan; cloud sessions work from GitHub repositories
GitHub Copilot cloud agentGitHub Actions environment or cloud sandboxGitHub only59 minute cap per Actions session; one pull request per task
Cursor cloud agentsIsolated cloud virtual machinesGitHub, GitLab, Bitbucket Cloud, Azure DevOpsNot on Free or Plus plans; billed against usage at model rates

OpenAI Codex

Codex is the closest in shape to Jules. OpenAI's documentation describes handing it coding tasks to work through in the cloud, with each task getting its own workspace that keeps running while your computer is off or asleep. It connects to GitHub or GitLab, lets you commit or open a pull request when the work is ready, and supports reusable environments with an install script, which is where exported Jules setup scripts belong. Network access is configurable, from package manager domains only to full public internet. The trade-off is that Codex is reached through ChatGPT accounts, so the cost is set by ChatGPT pricing rather than by a coding product price, and teams already standardised on another assistant end up paying for two.

Claude Code

Claude Code started as a terminal agent and now also runs tasks in cloud sessions: you pick a GitHub repository, describe the task, and it works in an isolated virtual machine and pushes its changes to a branch for review, continuing if you close the page. Anthropic lists cloud sessions on its Pro, Max, Team and Enterprise plans. The strength is that the same agent works locally and remotely, so a team can run it on a laptop for interactive work and in the cloud for the backlog work Jules handled. The trade-off is that it is sold through Claude subscriptions rather than as a separately metered coding service, so check whether your plan's usage limits suit sustained unattended work before moving scheduled jobs onto it. See Claude Code pricing and our Claude Code vs GitHub Copilot comparison.

GitHub Copilot cloud agent

For teams whose code is entirely on GitHub, the agent built into GitHub Copilot is the least disruptive replacement, because it lives where Jules's output already went. GitHub's documentation says it can research a repository, plan, edit files, run tests and linters, and open a pull request, and from 28 October 2026 the new cloud experience is available to users on any paid Copilot plan. The limits are stated plainly and worth reading before you migrate scheduled work: each session running on GitHub Actions has a maximum execution time of 59 minutes that cannot be extended, the agent works on one branch at a time and opens one pull request per task, it only works with repositories hosted on GitHub, and branch protection rules can block it. Compute draws on the GitHub account's monthly allowance, separately from AI credits. See GitHub Copilot pricing and the GitHub Copilot alternatives if the GitHub-only restriction rules it out.

Cursor cloud agents

What used to be called background agents in Cursor are now cloud agents, which run "in isolated VMs in the cloud with full development environments" and do not need your machine to stay online. They have the broadest source control support here, covering GitHub, GitLab, Bitbucket Cloud and Azure DevOps, and the most launch points: the desktop app, the web, iOS, Slack, Linear, GitHub comments and an API. That makes Cursor the most direct replacement for a Jules API integration. The caveats are cost and ownership. Cloud agents require a Premium, Teams, Enterprise or legacy paid plan, are excluded from Free and Plus, and bill against usage at the selected model's rate, with a spend limit requested on first use. And Cursor is now owned by SpaceX, which we covered in what the SpaceX acquisition means for Cursor users. Compare it directly in Cursor vs GitHub Copilot.

And the ones we left out

Devin, Windsurf and several open source agents also run coding tasks. The five above are simply the closest match to what Jules users depended on: a hosted or Google-backed agent and a GitHub pull request at the end. The wider field is in our developer tools category.

How to choose, and what this says about AI tools

The choice comes down to three questions, in this order.

Where is your code?

If everything is on GitHub and you already pay for Copilot, start with the Copilot cloud agent: it costs the least to adopt and its limits are documented. If you also use GitLab, Bitbucket or Azure DevOps, Cursor and Codex are the realistic options, since the Copilot agent is GitHub-only.

Did Jules run unattended?

If Jules ran scheduled tasks or was triggered from CI, you need a hosted agent with an API or a trigger, which means Codex, Cursor cloud agents, Claude Code cloud sessions or Copilot. If you only used it interactively, a local agent such as Antigravity or Claude Code in the terminal is enough, and it is cheaper to run.

Which assistant does the organisation already pay for?

The cheapest replacement is usually the one already on the invoice. Adding a fifth AI subscription to replace a tool that was bundled with Google AI Pro is the kind of quiet spend growth we described in the AI tier you did not ask for.

The lesson that applies to the replacement too

Jules lasted roughly two years from its December 2024 debut, had passed 140,000 public commits by August 2025 according to Google's own changelog, and still closed with thirteen days of working notice. Every tool on the list above is younger than most of the code it will touch. Choose one, but keep the agent's configuration, setup scripts and instructions in your repository rather than only inside the vendor's product, so that the next migration is a configuration change and not a rebuild. That is the practical lesson of this shutdown, and of the others in our 2026 shutdowns roundup.

Questions people ask

When is Google shutting down Jules?

Jules winds down on 31 December 2026. Users can keep finishing sessions until 11 November 2026, after which Jules is read-only until the end of the year. Google announced the shutdown in the Jules changelog on 29 October 2026.

Will I lose my code when Jules shuts down?

No. Jules worked through GitHub, and Google states that code, branches and pull requests "already live in GitHub, so they stay where they are." What you can lose is the data held inside Jules: settings, environment variables, scheduled tasks, connected repositories, memories, setup scripts and session history. Export those before 31 December.

How do I export my Jules data?

Install Jules Tools with npm install -g @google/jules, run jules login, then open a command line coding agent in your home directory and paste the export prompt from the Jules changelog. The export includes a file of environment variable values that contains secrets, so keep it private and out of any repository.

Will scheduled tasks and the Jules API keep working until 31 December?

Plan as if they will not. Google says Jules turns read-only after 11 November 2026, and the notice does not separately address the API or scheduled tasks. Treat 11 November as the date any automation that creates Jules work stops, and replace or remove it before then.

Is Jules being replaced by Antigravity?

Google says the work is folding into Antigravity, its agentic development platform, but has not announced when Jules capabilities will be available there or whether Jules data will transfer. Antigravity's own page describes agents that run locally and is free for individual developers; it does not currently describe hosted asynchronous sessions like Jules.

Will Google refund Google AI Pro or Ultra subscribers?

The shutdown notice does not mention refunds or any change to Google AI Pro or Ultra, the plans that raised Jules task limits. Those plans include other Google AI products, so ask Google support directly and get the answer in writing before your next renewal.

What is the best alternative to Jules?

It depends on where your code lives and whether Jules ran unattended. For GitHub-only teams already paying for Copilot, the GitHub Copilot cloud agent is the least disruptive. For hosted agents across GitHub and GitLab, OpenAI Codex and Cursor cloud agents are closest. Claude Code suits teams that want one agent both locally and in the cloud. Antigravity is free and suits interactive, local use.

Sources

Every price, limit and date above was checked against these pages on the day shown. Where a figure has since moved, the vendor’s own page is the authority and this one is a snapshot.

  1. 1Jules changelog · GoogleThe 29 October "Jules Turndown" entry: the 11 November and 31 December dates, the export contents and method, and earlier entries for the API, CLI, scheduled tasks and plans.Checked
  2. 2Google Is Shutting Down AI Coding Agent Jules As It Moves Its Work To Antigravity · Times of AIIndependent report confirming the timeline, the December 2024 debut and that no Antigravity availability date has been given.Checked
  3. 3Google Antigravity · GoogleWhat Antigravity is: app, CLI, IDE and SDK, local agents, free for individual developers.Checked
  4. 4About Copilot cloud agent · GitHubThe 59 minute session cap, one pull request per task, GitHub-only scope and the 28 October availability for paid plans.Checked
  5. 5Cloud agents · CursorWhere Cursor cloud agents run, supported source control, launch points and plan requirements.Checked

Read next

  • News

    The 2026 Shutdown List: Software That Stopped, And What The Vendors Actually Said

    OpenAI shut the Sora app in April and switches the API off on 24 September. Foundry stopped developing Modo after seventeen years. Microsoft retired Teams live events in June. Alongside the well-covered closures, rechecking 2,898 products against their vendors’ own pages turned up a set of quieter ones. This is the list, with what each vendor said and where they said it.

  • Analysis

    The AI tier you did not ask for is the biggest line on your 2026 renewal

    Software spend rose nearly 8% in a year while the number of applications stayed flat, which means the same portfolio now costs considerably more. The mechanism is not inflation and it is not seat growth. It is the AI-enhanced tier that arrives at renewal as the only tier still being sold, and September 2026 gave buyers three dated examples of how the pattern works.