Sundar Pichai said Google is “a bit behind” the frontier in agentic coding, especially on tool use, instruction following and long-horizon work on complex codebases.
Google’s CEO tied that gap to a missing developer product surface: rivals had coding tools generating daily interaction data, while Google lacked comparable external feedback loops to improve models.
Antigravity 2.0 — unveiled at I/O as a standalone desktop coding app — is now meant to build that loop, with internal usage doubling every week, according to Pichai.
Pichai’s comments came a day after Google made Gemini 3.5 Flash the default model for AI Mode globally and acknowledged early complaints over pricing, quality and tight usage limits imposed to avoid outages.
Google says Gemini 3.5 Pro is already used internally and should roll out next month, but it has not said whether that release will close the coding gap.
Can Google's new agentic AI strategy help it reclaim dominance, or is it too little, too late in the AI race?
As AI agents learn to code autonomously, is the human developer's role shifting from creator to mere overseer?
Google Antigravity 2.0 Launch: The Agent-First IDE Revolutionizing AI-Powered Software Development
Overview
At I/O 2026, Google launched Antigravity 2.0, marking a major move into agentic coding. This came as developers had been turning to competitors like Claude Code and Codex, showing a strong demand for advanced AI-powered development tools. Google aims to strengthen its position by focusing on agentic coding, where autonomous AI agents can understand complex contexts and handle multi-step tasks. Antigravity 2.0 demonstrates this by building a simple operating system from scratch in just over 12 hours, highlighting Google’s commitment to leading in agentic workflows and meeting the evolving needs of developers.