Google's answer to the coding-agent race

Google Jules started as a Labs project, but has now exited beta and become more broadly available. That makes Jules one of the most relevant coding agents for developers already working within the Google ecosystem, Gemini, and GitHub.

Jules is described as an asynchronous coding agent. It connects to GitHub, clones the repository inside a virtual machine, draws up a plan, waits for approval, and gets to work while you do something else. When it's done, you receive a diff, an explanation, and the option to take the change further.

The most important aspect is the way it works: Jules isn't just trying to be a smarter autocomplete. It's trying to become a background worker.

Jules points toward a developer workflow where you don't ask AI about code — you delegate small jobs to dedicated work environments.

VMs and plan-first

In its documentation, Google explains that each task runs inside a virtual machine where Jules clones the code, installs dependencies, and makes changes. Users can add setup scripts so the project builds and tests correctly.

This matters in practice. Many agent failures stem from poor environments: missing env vars, the wrong Node version, incomplete install steps, or tests that never run. An agent that can reuse a previous setup and show its plan before touching code is far easier to keep under control.

Google also notes that Jules automatically looks for AGENTS.md in the repository root — a signal that agent instructions are becoming standard infrastructure.

Google Jules moves from beta to working agent for developers - Bilde 1

API and CLI make Jules more compelling

In October 2025, Google announced Jules Tools, a lightweight CLI, and the Jules API. The API lets developers programmatically create sessions, approve plans, and integrate the agent into their own workflows.

That opens up far more than manual use in a browser. A team can connect Jules to Slack, Linear, GitHub, or internal systems. A support bug can become a suggested fix. A routine chore can become a scheduled agent task.

But teams need to be disciplined here. API access to a coding agent is powerful and should be treated as a production integration — complete with logging, access controls, and clear stopping points.

3 | entry points | web | CLI | API

Relevance for developers

Jules is especially interesting for small teams with many small tasks but limited bandwidth. Product teams will recognise the pattern: the backlog is full of "should fix" items, but the sprint gets consumed by customer deliverables.

Jules can be useful for:

  • Documentation updates.
  • Simple bug fixes with solid reproduction steps.
  • Test generation around existing code.
  • Dependency upgrades in well-scoped packages.
  • First drafts of refactoring plans.

It should not be the first choice for large architectural changes, security-critical code, or database migrations without a human-authored plan.

The clearer the task, the more Jules becomes a working agent. The vaguer the task, the more it becomes an expensive guessing machine.

Pricing and access matter

Google has introduced structured Jules tiers, including higher limits for Google AI Pro and Ultra. The documentation also indicates that paid plans are tied to Google AI Plans for individual Google accounts, with further work planned for other user groups.

For organisations, this means Jules may be easier to trial individually than to roll out company-wide. That makes the pilot setup especially important: validate the value before building an internal process around the tool.

Conclusion

Jules is a clear sign that coding agents are becoming workflow products. Google combines Gemini, GitHub integration, VM environments, plan approval, CLI, and API into a single package.

The smart starting point is straightforward: connect Jules to one safe repository, let it work through small issues, and review the results as pull requests from a new colleague. Don't measure magic. Measure whether the backlog actually moves.