Google opens the Jules coding agent in public beta

Background

A coding agent is a system that takes a described task — fix this bug, add that feature — and carries it out in a codebase itself, rather than suggesting the next line while a programmer types. The word "asynchronous" describes when the work happens: the agent runs in the background, so a developer can hand off a task, do something else, and come back later to review what came out. Agents of this kind need access to the place where code actually lives, which in most teams is a repository hosted on GitHub, together with the review and issue-tracking tools built around it.

The launch

On 20 May 2025 Google opened Jules, its asynchronous coding agent, to everyone in a public beta^[1]. Opening it to everyone meant that any developer could try the agent without being part of a waitlist or a private programme.

How it works

Jules works with a user's GitHub repositories. To do that, it clones the code into a secure Google Cloud VM^[1] — a copy of the repository is placed on an isolated machine in Google's cloud, so the agent's edits, builds and test runs happen away from the developer's own computer and away from the live copy of the project. Google said the agent works directly inside users' GitHub workflow^[1], meaning tasks are set up and results handed back through the same tools developers already use to manage their code, rather than in a separate environment they have to learn.

The model behind Jules was Gemini 2.5 Pro^[1]. According to Google, the agent can write tests, build features, fix bugs, bump dependency versions and provide audio changelogs^[1]. Those tasks cover much of ordinary maintenance work. Tests are the checks that code behaves as intended, and writing them is often skipped when a deadline is close. A feature is a new piece of behaviour in the product. A bug fix corrects something that does not work. Bumping a dependency version means moving a project onto a newer release of an outside library it relies on — routine, necessary for security and compatibility, and tedious to do by hand across many packages. The audio changelog is a spoken summary of what the agent changed, which is meant to make it quicker to take in a batch of edits than reading a diff line by line.

Pricing

The beta was free of charge, with usage limits^[1]. Google said it expected to introduce pricing for Jules after the beta, as the platform matured^[1]. That is the usual shape of a preview release: wide, free access at first to gather feedback and usage patterns, with a paid tier added once the product and its cost structure are better understood.

Why it matters

The open beta put a general-purpose coding agent in front of anyone who wanted to try one. That is a different proposition from the autocomplete-style assistants that had become common in code editors. Autocomplete helps with the line being written; an agent is handed a task and returns a change. The practical questions shift accordingly — towards whether the change is correct, whether the tests it wrote actually test the right thing, and how much of the result a reviewer has to check by hand. Because Jules works inside the GitHub workflow and returns its work there, review happens in the same place a team already reviews code, which is what makes the hand-off usable rather than a separate chore. The stated plan to charge after the beta also marked the point at which such agents were being treated as products with a cost, not only as demonstrations.