Jules - An Asynchronous Coding Agent
Jules is Google's coding agent for delegating tasks in a GitHub repository. The official page names bug fixes, version updates, tests and feature work as intended tasks. Its workflow separates the request, plan, proposed code changes and pull-request review.
Giving a repository task
Select the GitHub repository and branch, then write a detailed prompt. The page also describes using a jules label on a GitHub issue to assign work. Include the intended behavior and acceptance checks so the agent has more than a vague instruction to improve the project.
Jules fetches the repository and clones it into a cloud virtual machine. That is a remote working environment, not a local terminal session on the developer's machine. For a project-directory agent with local-model configuration, Little-Coder offers a useful comparison; the execution environment is an important difference to evaluate.
Plan, diff and pull request
The depicted workflow has Jules develop a plan before changing the project. It then provides a diff for inspection and produces a pull request. The page leaves approval and merging to the developer.
Use those review points to compare the work with the original request. A dependency update can affect configuration and runtime behavior beyond the version number, while a passing generated test may not cover the relevant user path. Inspect the diff and run the project's own checks before accepting the change.
The publisher's example depicts a framework migration, but that is an illustration of the workflow, not proof that Jules can complete every migration without intervention. Prefer a bounded task with an observable result when evaluating it on an unfamiliar codebase.
Plan access and operational limits
The page distinguishes Jules, Jules in Pro and Jules in Ultra by task allowance, concurrency and model access. Its cards do not give identical model access to every tier. Check the current plan before scheduling parallel work rather than interpreting an overview's model language as a universal entitlement.
No fixed price, latest-model assertion or enduring quota is claimed here. Those details can change, and the captured overview is not a full security or data-retention specification.
Before connecting a private repository, review the permissions and service terms for the cloud workflow. Jules can prepare a change, but repository owners still decide whether the code, tests and resulting pull request satisfy the project.
