Google I/O 2026 doubles down on agents

Google looks more credible than noisy here: the model stack and agent tooling are getting real, but the company still ships them across too many surfaces for the experience to feel unified.

• –Gemini 3.5 Flash is the headline win, especially if it really holds up on coding and agentic workflows as Google claims
• –Gemini Omni and Spark push Google toward multimodal, proactive assistants instead of just chatbots
• –Antigravity and managed agents are the most developer-relevant pieces because they turn Google’s AI into an actual agent platform
• –The criticism about clutter is fair: Gemini, AI Studio, Search, Workspace, and standalone tools still feel fragmented
• –Usage limits and tiering remain the biggest practical downside; capability matters less if access feels constrained
• –The lack of a clear long-running “deep think” benchmark or flagship pro model update leaves part of the story unresolved