Google repositions Gemini at I/O as an agent platform for long‑horizon, multi‑step tasks

Google used its May 19 I/O 2026 conference to reposition Gemini around agents that can carry out longer, multi-step jobs across tools, rather than around a single flagship model release. The company introduced Gemini 3.5 Flash as the opening model in its 3.5 series and paired it with updates to Google Antigravity, its agent-focused development platform. Google said the combination is aimed at developers building coding tools, managed agents and enterprise workflows. Forbes and other outlets described the event as a broader platform push around agentic software rather than a narrow model announcement. ### Why did Google frame Gemini around agents instead of a single model? Google said in its May 19 product post that Gemini 3.5 combines “frontier intelligence with action,” language that puts execution alongside reasoning. The company said Gemini 3.5 Flash was designed to help users execute “complex, agentic workflows,” and Google Cloud said the model excels at “complex long-horizon tasks” for agents and coding. (blog.google) Sundar Pichai, Google’s chief executive, used the I/O keynote to describe products that can act in the background through what he called the Google Antigravity harness. In Google’s keynote recap, the company said Gemini Spark is powered by Gemini 3.5 and that harness, allowing it to perform long-horizon tasks and connect to tools, with third-party integrations through MCP coming in the following weeks. (blog.google) ### What exactly launched with Gemini 3.5 Flash? Google said Gemini 3.5 Flash is the first release in the Gemini 3.5 family. In its developer highlights post, Google said the model is available through the Gemini API and positioned it as the “high-speed engine” for real-world agentic workflows. (blog.google) Google Cloud said the 3.5 series began with Flash and described it as delivering frontier performance for agents and coding. Outside coverage added that Google presented Flash as the default fast model for agentic work, with several reports noting the emphasis on speed, coding and tool use rather than on a single benchmark race. (blog.google) ### Where does Antigravity fit in the new setup? Google said at I/O that Antigravity is its “agent-first development platform,” and the company’s I/O roundup said it had moved “beyond AI tools that just help us write, to agents that help us act.” In Google’s developer materials, Antigravity is presented as the environment around the model layer, not just a coding assistant. (cloud.google.com) Google’s May 21 developer highlights said Antigravity now sits alongside the Gemini API and AI Studio in the company’s developer stack. Third-party coverage from Forbes and others said the platform now includes enterprise and developer surfaces that make Gemini look more like a system for orchestrating agents across tasks and workspaces. (blog.google) ### What do the CLI and SDK tell developers about Google’s priorities? MarkTechPost and The Next Web reported that Antigravity now includes a CLI for terminal-based developers and an SDK for custom agent behavior and orchestration. Those reports said teams can define how agents behave and choose where to host them, pointing to a product pitch centered on control as much as model access. (blog.google) Google’s own posts do not use the same outside phrasing on governance, but they do show the same direction. The company’s materials describe Antigravity as a platform for independent agents and place it next to the Gemini API, AI Studio and Google Cloud, suggesting Google wants developers to build, route and run agent systems across multiple environments. That is an inference from Google’s published product structure and launch materials. (thenextweb.com) ### What should readers watch next? Google said third-party tool connections for Gemini Spark through MCP are due in the coming weeks. The company also said Gemini 3.5 Flash is already available through the Gemini API, while the broader Antigravity and Google Cloud stack gives developers the channels through which this agent strategy will be tested with customers. (blog.google 1) (blog.google 2)