Google unveils Jitro
Google introduced Jitro, an AI system described as capable of self‑coding and autonomous improvement, marking a visible shift toward models that can modify their own code. The announcement frames a technical pivot where models are built to iterate on their development rather than rely solely on human updates. (x.com/RoundtableSpace/status/2043082914988999151)
Google has not published a standalone announcement for “Jitro” on its official channels, but Google’s coding products have moved in that direction with Jules and Antigravity, both built around autonomous code work. (blog.google) In December 2024, Google said Gemini 2.0 was built for an “agentic era,” meaning models that can use tools, plan across steps and take actions with supervision. In that same launch, Google introduced Jules as an AI code agent for developers. (blog.google) By June 2025, Google described Jules as “Google’s autonomous AI coding agent” and moved it into public beta. The product was pitched as more than autocomplete: it reads a repository, proposes changes, and works asynchronously on tasks like bug fixes and tests. (blog.google) Google’s own Jules site now calls it a “proactive coding agent” that can “delegate your backlog,” “set up automated tasks,” and “proactively find and fix code improvements.” That language is closer to a system that keeps working on a codebase than a chatbot waiting for the next prompt. (jules.google.com) In November 2025, Google widened that approach with Antigravity, a development platform it said gives developers an “agent-first interface” where agents “autonomously plan, execute and verify complex tasks.” Google said Gemini 3 was adding “advanced agentic coding capabilities” across that stack. (blog.google) That is the technical backdrop for the “self-coding” claim around Jitro. In plain terms, the idea is software that does not just write a requested function, but inspects a larger codebase, chooses tasks, edits files, runs checks and proposes the next round of changes. (blog.google) Google has already shown pieces of that workflow in production. In a December 2025 update, the company said the team behind Stitch configured a daily “pod” of Jules agents for jobs including performance tuning, security patching, accessibility improvements and test coverage, and said Jules became one of the largest contributors to the Stitch repository. (blog.google) Google has also said Jules was out of beta by August 6, 2025, powered by Gemini 2.5, after developers used it on tens of thousands of tasks and shared more than 140,000 code improvements publicly. Those numbers suggest Google has been moving from code suggestions toward longer-running software agents for at least the past year. (blog.google) What remains unverified is the exact status of “Jitro” itself. Search results and social posts describe it as a next-generation Jules system, but Google has not posted a product page, research paper or formal blog post under that name that I could verify. (x.com) (devops.com) So the clearest reading today is narrower than the headline: Google has officially documented a steady shift from coding assistants to autonomous coding agents, and “Jitro” appears to be the label circulating for the next step in that line. Until Google publishes a primary-source announcement under that name, the confirmed story is the direction of travel, not the full product dossier. (blog.google 1) (blog.google 2)