Google opens asynchronous coding agent "Jules" to public beta
On May 20, 2025, timed with Google I/O 2025, Google launched its asynchronous AI coding agent “Jules” into public beta. Having first appeared as an experimental Google Labs project for a limited group of testers in December 2024, it entered a simultaneous worldwide beta roughly five months later. Built on Gemini 2.5 Pro, its defining trait is autonomously processing code on cloud virtual machines.
Details
- Asynchronous agent: Unlike a traditional “copilot” style tool, it’s designed to read code, understand intent, and work through tasks in the background while the human works on something else
- Works with real codebases: Operates with the context of an entire real project rather than a sandbox, and can process multiple tasks in parallel on secure virtual machines on Google Cloud
- Transparency: Presents its plan and reasoning before starting implementation, allowing users to revise the plan before, during, and after execution
- GitHub integration: Plugs into existing GitHub workflows to write tests, add features, fix bugs, and update dependencies
- Audio changelog: Also includes an “audio changelog” feature that lets users listen to a summary of changes
- Terms of availability: Free with usage caps during the beta period, with no waitlist required to start using it immediately; pricing was said to be introduced after the beta ends
What happened next
During the public beta, thousands of developers tackled tens of thousands of tasks, and more than 140,000 code improvements were shared publicly, generating significant buzz. Building on that track record, Jules moved to general availability (GA) on August 6, 2025, relaunching as a full-fledged product with free, Pro, and Ultra plan tiers.