That backlog piling up in your issue tracker — when are you going to clear it? Now you can just throw it over the wall. Assign a GitHub issue, and Jules reads the code, makes a plan, finishes multi-file changes, and hands back a PR.
What is this?
Jules is Google's asynchronous autonomous coding agent. While existing AI coding tools were "copilots suggesting code alongside you," Jules is closer to a pilot that independently performs tasks. Connect it to a GitHub repo, and it takes issues, analyzes the codebase, shows you a modification plan, and once approved, executes in a cloud VM and returns the results as a PR.
It started as a public beta in May 2025, then added a key feature called Suggested Tasks in December 2025. It automatically scans repos for #TODO comments and performance improvement opportunities, then proactively asks "should I fix this?" It finds work to do without being asked.
Then in November 2025, Gemini 3 Pro was integrated into Jules. Compared to the previous model (Gemini 2.5 Pro), reasoning is much more stable on multi-step tasks, and it grasps developer intent more accurately. In Google's official blog's words, "multi-step tasks flow much more naturally."
Key analogy
GitHub Copilot = a copilot suggesting code beside you
Jules = a junior developer who takes an issue assignment and comes back with a PR on their own
What changes?
The AI coding tool market is heating up, but each tool has a different position. Jules' differentiator is being asynchronous + proactive. Other tools only move when you ask for something, but Jules monitors repos and finds work to do on its own.
| GitHub Copilot | Claude Code | Google Jules | |
|---|---|---|---|
| How it works | IDE inline autocomplete | Terminal-based agent | Async cloud agent |
| Trigger | Auto-suggests while coding | Explicit commands | Issue assignment or auto-detection |
| Multi-file edits | Limited | Powerful (200K context) | Powerful (cloud VM) |
| Proactive suggestions | None | None | TODO scanning, performance optimization suggestions |
| Output | Code snippets | Local file changes / PR diff | Auto PR generation |
| Cost | $19/mo+ | Token-based ($20/mo+) | Included with AI Ultra subscription |
With Gemini 3 Pro onboard, a few things specifically improved:
- Enhanced Critic Agent — The "critic" agent that re-reviews its own code mid-task now works much more reliably. It intervenes again after replanning to maintain quality.
- Intent Alignment — When you say "refactor this function," it touches only the requested scope with no side effects. Much more precise than before.
- Safer Git handling — Reduced risk of force pushes or unintended branch manipulation.
The essentials: how to get started
- Connect Jules
Log in with your Google account at jules.google.com → connect your GitHub repo. Requires a Google AI Pro or Ultra subscription. - Throw your first task
Assign Jules to a GitHub Issue in your connected repo, or create a task directly in the Jules web UI like "fix this bug." Jules will show you a plan — review it and approve. - Install Jules Tools CLI
Install withnpm install -g @google/jules. Runjules statusat your project root and it automatically finds the.git folder to recognize the repo. Usejules diffto check changes andjules remote list --taskto see tasks in progress. - Delegate backlog in parallel
You can pipe TODO files or GitHub issues to create tasks in bulk. Likecat TODO.md | jules remote create. Use the--parallelflag to run multiple tasks simultaneously, cutting execution time by 50%+. - Turn on Suggested Tasks
Enable Suggested Tasks for up to 5 repos in Jules settings. Jules automatically finds #TODO comments and performance optimization points to suggest. Approve what you like, dismiss what you don't.
Good to know
Jules still has context window limitations with large codebases. Rather than trying to process an entire monorepo at once, delegating clearly scoped issues has a much higher success rate. Also, the free tier has tight daily usage limits, so consider AI Pro ($19.99/mo) or higher for serious use.
