Jules

Filed by the Vibe Code Weekly desk · Updated Nov 12, 2026

The story

Jules is Google's asynchronous coding agent. A developer hands it a job against a GitHub repository, it does the work on a cloud virtual machine, and it returns a pull request. It sits with the delegate-then-review tools in AI coding, not with the autocomplete helpers that live inside an editor.

AI coding products tend to fall into two groups. Some sit in the editor and suggest lines as you type. Others take a whole job and come back with finished work. Jules belongs to the second group. Google aims it at the chores developers tend to postpone, such as fixing bugs, bumping versions, writing tests and building features. The stated goal is more time for the code people actually want to write.

There are two ways to start a job. In the Jules interface, you pick a repository and a branch and type a detailed request. Inside GitHub, you can tag an issue with the "jules" label to assign it to the agent. Jules then copies the repository onto a Cloud VM and proposes a plan, which you can accept. In Google's demo, the resulting diff updates package.json to move the next dependency from 10.2.3 to 15.4.5.

Finished work comes back as a diff you can browse and approve, then as a pull request that you merge and publish yourself. Because the work runs on remote machines rather than a laptop, several jobs can run side by side. The three tiers differ in daily task allowance, how many tasks can run at once, and which Gemini models you get. For people who adopt it, more of the day goes to reviewing proposed changes and less to typing them.

Jules is Google's asynchronous coding agent. A developer hands it a job against a GitHub repository, it works on a cloud VM, and it returns a pull request. Google aims it at routine chores such as bug fixes, version bumps, tests and feature building.

The good

Work can be assigned from inside GitHub with an issue label. The agent proposes a plan that you can approve. Results arrive as a diff and a pull request, which fits the review habits teams already have. Jobs run on a Cloud VM, so several can run in parallel. Higher tiers include newer Gemini models, starting with Gemini 3 Pro.

The catch

The workflow assumes GitHub repositories and branches. The entry tier allows 15 tasks a day, with 3 running at once. The entry tier runs on Gemini 2.5 Pro, and newer models need a higher plan. Someone still has to read the changes and merge them.

Who should read on

Teams whose code already lives on GitHub and who want to offload routine maintenance. Turning labeled GitHub issues into proposed fixes. Framework and dependency upgrades that touch many files. Developers who want several background jobs running at once.

Who can skip it

Anyone without a GitHub repository to point it at. People who expect a live preview or drag-and-drop editing instead of code review.

Reader mail