Meet Jules Tools: A Command Line Companion for Google's Async Coding Agent

A Google for Developers post, written by two engineers on the Google Labs team, announcing that Google Jules, Google’s asynchronous coding agent, can now be driven from the command line through a new CLI called Jules Tools. The stated motivation is that developers had until then interacted with Jules primarily in a web browser, while — in the authors’ words — developers live in the terminal, where they test, build, debug and ship.

The post’s argument is that a command line surface changes what the agent is rather than only where it is reached from: it makes Jules programmable, scriptable and customizable, so it can be wired into a developer’s own automations rather than operated as a separate destination. The authors describe the result as both a dashboard and a command surface for a coding agent, and argue in a “Looking ahead” section that the future of development tools is hybrid.

Key Points

• The post describes Jules as an asynchronous coding agent that integrates with existing repositories, understands a project’s full context, and performs tasks such as writing tests, building features, providing audio changelogs, fixing bugs and bumping dependency versions.
• It states how a task runs: Jules spins up a temporary VM, does the work there, and sends back a pull request, with nothing running until the developer asks it to.
• The CLI installs with npm install -g @google/jules and is built around commands, which tell Jules what to do, and flags, which adjust its behaviour — the post’s examples are jules remote list --task and jules --theme light.
• Because it is scriptable, the post shows it composed with other command line tools: listing connected repositories, creating a remote session against a named repository, looping a TODO file into multiple sessions, piping a GitHub issue title from gh into a new session, and using the Gemini CLI to pick the most tedious issue and hand it to Jules.
• Jules Tools also offers a TUI for interactive use, where commands such as /remote give a dashboard view of tasks and /new walks through creating one — which the post presents as the same control as the web UI, faster and closer to where the developer works.
• The authors state a “hybrid by design” position along two axes: local plus remote — use your own machine when you want, spin up multiple VMs when you need scale — and do-it-yourself plus delegation, staying hands-on with code while offloading work to the agent.

Context

This is a first-party product announcement written by two of the product’s own staff, so its claims about what Jules does and why a terminal surface matters are the vendor’s own. It reports no usage data or evaluation.

The post’s “hybrid by design” framing — keeping the developer hands-on while delegating in parallel — is a position about how much autonomy to hand an agent, adjacent to Human-in-the-loop and the Supervised Agency Spectrum. Its emphasis on making the agent scriptable and composable with existing CLI tools places it alongside other work on the Outer Loop and on Long-running Agent workflows.