Google Jules

Jules is Google's asynchronous coding agent, built in Google Labs and powered by the Gemini family of models. It belongs to a different category than an in-editor assistant: rather than sitting beside you and suggesting code as you type, Jules takes a complete task, usually a GitHub issue, and goes off to do it on its own while you work on something else. When it is finished, it hands back a pull request for review.

The mechanics are what make it distinct. Given a task, Jules clones your repository into a secure, isolated Google Cloud virtual machine, reads the full project context, and writes a step-by-step plan that you can review, edit, or reject before any code is written, so you steer the approach rather than just approving the result. It then makes the changes across multiple files, runs your existing test suite, and iterates when tests fail. Because the VM has network access, Jules can install dependencies and run real builds. When the work is done it pushes to a branch and opens a pull request with clear commit messages and a plain-language summary, and if your CI pipeline fails on that PR, Jules can read the error, apply a fix, and push again on its own.

The real payoff is parallelism. Each task runs in its own cloud machine, so you can queue many at once and let them proceed in the background instead of blocking your editor. Jules runs on Gemini, using stronger models for planning and lighter ones for simpler subtasks, and supports common languages including Python, TypeScript, JavaScript, Go, Rust, and Java. It integrates deeply with GitHub, can be driven from a command-line tool and a public API to wire it into CI pipelines, Slack, Linear, and Jira, and remembers per-repository preferences across tasks.

Google keeps Jules private by default, isolating code in ephemeral VMs and stating it does not train on private code. It is best suited to well-specified work such as bug fixes, dependency upgrades, test writing, and batched refactors, where a solid brief and good test coverage let an agent run unattended, and it complements interactive tools used for real-time exploration.