Jules: How to Use Google's Asynchronous Coding Agent
Jules is an asynchronous coding agent from Google Labs. Its biggest selling point is that you can hand off a task and stop watching it. Jules copies your entire
Jules is an asynchronous coding agent from Google Labs. Its biggest selling point is that you can hand off a task and stop watching it. Jules copies your entire GitHub repository into an isolated virtual machine on Google Cloud, where it reads the code, drafts a plan, edits files and runs tests. When it finishes, it packages the result as a Pull Request. It works best for tasks with a clear scope and clear acceptance criteria, such as upgrading package versions, adding unit tests or fixing known bugs. It is a poor fit for development where requirements are still vague and need to be worked out as you code. What Is Jules: From Experimental Project to Official Product Jules first appeared in December 2024 as an experimental coding agent. It was one of the projects Google used to show off the agent capabilities of Gemini 2.0, and at that point only a small group of testers could use it. For the full announcement, see the Gemini 2.0 developer announcement on the Google Developers Blog . At Google I/O in May 2025, Jules moved to a global public beta and switched to Gemini 2.5 Pro. The beta's results can be measured: “During the public beta, developers completed tens of thousands of tasks with Jules and publicly shared more than 140,000 code improvements (August 2025)” (Source: Google official blog) . In August of that year, Jules left beta and launched tiered plans. The model underneath Jules sets the upper limit on how complex a change it can handle. “Gemini 2.5 Pro scored 63.8% on the SWE-bench Verified software engineering benchmark (March 2025, using a custom agent setup)” (Source: Google DeepMind official blog) . This benchmark uses real GitHub issues to test whether a model can produce patches that pass the tests, which closely matches how Jules is used in practice. What “Asynchronous” Means: How Jules Differs from Synchronous AI Assistants Jules and in-editor autocomplete tools work in different ways. GitHub Copilot's inline suggestions and Cursor's live chat are
FAQ
What Is Jules: From Experimental Project to Official Product
Jules first appeared in December 2024 as an experimental coding agent. It was one of the projects Google used to show off the agent capabilities of Gemini 2.0, and at that point only a small group of testers could use it. For the full announcement, see the Gemini 2.0 developer announcement on the Google Developers Blog . At Google I/O in May 2025, Jules moved to a global public beta and switched to Gemini 2.5 Pro. The beta's results can be measured: “During the public beta, developers completed
What “Asynchronous” Means: How Jules Differs from Synchronous AI Assistants
Jules and in-editor autocomplete tools work in different ways. GitHub Copilot's inline suggestions and Cursor's live chat are synchronous: the developer stays in the editor and approves each step. Jules runs on its own in the cloud once you hand off a task. You can close the browser, work on something else and come back to review the result when the task is done. How Jules Runs a Task Connect a repository : Authorize with your GitHub account and choose the repo and branch Jules can access. Clone
Plans and Quotas: How Many Tasks Can the Free Plan Run per Day?
Jules plans differ in two limits: tasks per day and concurrent tasks. According to “The Jules free plan offers 15 tasks per day with 3 running concurrently (August 2025 general availability announcement)” (Source: Google official blog) , paid plans come with Google AI subscriptions and raise these limits: Free plan : 15 tasks per day, 3 at a time. Good for personal projects and trying it out. Google AI Pro : About 100 tasks per day, about 15 at a time. Good for individual developers who code eve
Jules's Limitations and When Not to Use It
Jules's most obvious limitation is its slow feedback loop. Going from cloning the repository to producing a diff usually takes anywhere from a few minutes to several tens of minutes. For exploratory work where you want to try different approaches quickly, a synchronous in-editor assistant is more efficient. New features with unclear requirements : If the product spec is still under discussion, Jules can only follow your words literally, and its output often needs heavy rework. Projects that depe
Related Guidebooks
Reviewed and verified by FeiYueh · Last verified 2026-10-15. Independently maintained — not AI-generated boilerplate.
← Back to Blog