Google launches Jules AI agent for programming
Google has officially introduced the AI programming agent Jules, ending the beta testing that started in May. The tool is built on the Gemini 2.5 Pro model and allows you to run tasks asynchronously, works with GitHub, automatically creates environments on Google Cloud virtual machines, and is capable of updating code in real time without the involvement of a developer. About this informs TechCrunch.
The project was announced back in December as a Google Labs experiment and since then has received hundreds of improvements to the interface and functionality. According to Google Labs director of product Katie Korevec, it was the increase in stability that was the deciding factor in completing the test.
Since the release, Google has also introduced a new pricing grid. The free plan allows you to run up to 15 tasks per day, of which no more than three can be run simultaneously. For users who need advanced functionality, there are paid AI Pro and Ultra packages for \$19.99 and \$124.99, respectively, with 5x and 20x higher limits.
The company has updated its privacy policy: only public repositories can be used to train models, while private repositories remain protected. After exiting the beta, Jules got more opportunities to work with GitHub — now he can open not only pull requests, but also branches. The new Environment Snapshots feature allows you to save dependencies and scripts, which speeds up the execution of tasks.
During testing, users ran tens of thousands of tasks and shared more than 140,000 public code improvements. Jules learned to work even with empty repositories, which made the tool more accessible. According to SimilarWeb, the platform has been visited by 2.28 million users since May, of which 45% came from mobile devices. Most of the traffic came from India, the USA and Vietnam.
Although there is currently no mobile app, Jules is heavily used via the web interface, and Google is considering expanding support for smartphones. Within the company, the number of internal projects that have already implemented this tool is also growing.




