Jules

Jules is an autonomous coding agent from Google. The provider describes it as a tool that takes over the programming tasks you would rather not do yourself: fixing bugs, updating dependencies and versions, writing and running tests, adding documentation, or building new features. Unlike an assistant that thinks along in your editor line by line, Jules works asynchronously. You hand over a task, get on with something else, and receive a result to review later.

How Jules works

You start by signing in with your Google account at jules.google.com. Then you give Jules access to your GitHub repositories, pick a repository and a branch, and describe the task in plain sentences. Before anything is changed, Jules drafts a plan that you can approve or adjust. Only then does it start implementing. Tasks can also be triggered straight from GitHub by giving an issue the label "jules", which suits teams that already organize their work through issues.

A cloud VM instead of your own machine

According to the documentation, every task runs in a fresh, short-lived virtual machine in the cloud. There Jules clones the repository, installs dependencies, and runs tests. Common stacks such as Node.js, Python, Go, Java, Rust, and C/C++ come preinstalled, along with tools like Docker and Git. For more complex projects you can store a setup script and verify it with "Run and Snapshot" so later tasks start faster. The provider also points out that you remain responsible for the code you let it run, so you should review its changes like any other pull request before adopting them.

AGENTS.md and project context

Jules looks for a file called AGENTS.md in the root of your repository and reads hints about structure, conventions, and tooling from it. If you already use Claude Code, Codex, or other agents, you can reuse the same file and avoid maintaining a second one. The more clearly it describes how to build, test, and format the project, the fewer questions Jules has to ask.

Jules Tools: the command line

Besides the web interface there is the Jules Tools command line, which you install globally with npm. After signing in with your Google account you can create and list sessions and pull results, including several sessions in parallel. Run without arguments, it opens a terminal dashboard with session management and a diff view. That makes it possible to wire Jules into scripts and your own workflows.

Privacy

According to the documentation, Jules does not train on the contents of private repositories. For teams working with company code that is an important statement, but it does not replace your own review of Google's privacy policy and terms, which are linked on the website.

Plans

There are three tiers. The base tier is meant for trying things out and can be used free of charge, with a limited number of daily tasks and only a few running at once. The Pro and Ultra tiers raise those limits considerably and, according to the provider, are tied to the Google AI subscriptions; the higher tiers also give access to newer models. Current limits and terms are listed on the provider's site.

Who Jules is for

Jules suits developers with many small, clearly scoped jobs: dependency updates, missing tests, cleanup work, or well-described bugs. It is less suited to open-ended architecture questions or work that needs constant back-and-forth, where interactive assistants such as Cursor, Zed, or Warp are the better choice. Jules pays off most when the repository has solid tests and a well-kept AGENTS.md, because the agent can then verify for itself whether its change works.