Google unveils self‑coding AI
Google launched Jitro, described as an AI that can write code for itself, signaling another step toward autonomous model improvement. The announcement positions Google in the growing field of agents and tools that aim to reduce manual development cycles. (x.com)
Google is pushing deeper into software that writes software, extending a coding push that already includes Jules, Gemini Command Line Interface and a newer Google AI Studio agent. (blog.google) A coding agent is different from autocomplete: it can read a repository, plan changes across files, run tasks in the background and return a diff for review. Google described Jules that way when it opened public beta on May 20, 2025. (blog.google) Google said Jules clones a codebase into a secure Google Cloud virtual machine, works asynchronously, and shows its plan and reasoning before applying changes. The company said private code stays isolated in that execution environment and is not used for training by default. (blog.google) By August 6, 2025, Google moved Jules out of beta and said developers had shared more than 140,000 code improvements during testing. Google also tied higher usage limits to its Google AI Pro and Google AI Ultra subscriptions. (blog.google) Google broadened that agent strategy again on March 18, 2026, when it added the Antigravity coding agent to Google AI Studio. Google said the tool can turn prompts into web apps, add databases and authentication, and connect to outside services through stored application programming interface keys. (blog.google) The same shift is visible in Google’s developer tools. Google’s Gemini Command Line Interface documentation says the open-source terminal agent uses a “reason and act” loop with local tools and Model Context Protocol servers to fix bugs, add features and improve test coverage. (developers.google.com) Inside Google, the company says these tools are already affecting how engineering work gets done. In an August 18, 2025 post, Google said artificial intelligence was generating 30% of new code at the company and had lifted engineering velocity by an estimated 10%. (blog.google) Google is not alone in that race. OpenAI introduced Codex on May 16, 2025 as a cloud software-engineering agent that can work on many tasks in parallel, and Anthropic says Claude Code can read a codebase, make changes across files, run tests and deliver committed code. (openai.com) (anthropic.com) Google has not published a primary-source announcement about “Jitro” on its official sites that was visible in this search, so the clearest verified story is the direction of travel: Google has been moving from code suggestions to agents that plan, execute and hand back software changes for humans to approve. (blog.google)
