Google to push agent orchestration, may unveil next-gen Gemini at I/O

Google will open its I/O 2026 developer conference on May 19 with official event materials promising “latest Gemini model updates” and sessions centered on agentic workflows, coding tools and Chrome integrations. Google’s schedule lists a keynote at 10:00 a.m. Pacific time on May 19 and a developer keynote at 1:30 p.m., with AI and Chrome among the main tracks. Google has not publicly confirmed a product called “Gemini Spark,” but recent reports from Android-focused outlets said the company could show an always-on assistant designed to carry out multi-step tasks across apps. Google’s recent product posts and developer documentation show the company has already been building the plumbing for that kind of software, including Project Astra, MCP support for Google services and WebMCP for websites. ### What has Google itself said before I/O begins? Google said on February 17 that I/O 2026 would run May 19-20 at Shoreline Amphitheatre in Mountain View, California, and online. In a separate Google Developers Blog post the same day, the company said attendees should tune in for “agentic coding and the latest Gemini model updates.” Those official descriptions stop short of naming a new flagship Gemini release, but they place Gemini and agentic tooling at the center of the conference. (io.google) The I/O schedule now live on io.google adds more detail. Google lists sessions including “Agent-first workflows from prompt to production” and a Chrome session on using DevTools and MCP servers with coding agents such as Gemini CLI. Those listings indicate Google plans to spend conference time on how models connect to tools, browsers and execution environments, not only on model benchmarks. (blog.google) ### Where does the “agent orchestration” theme come from? Google Cloud used the term directly on April 22 when it rebranded and expanded Gemini Enterprise around “agent development, orchestration, and governance.” In that post, Google said companies need ways to build, scale, govern and optimize agents that can execute complex workflows, and said its platform includes support for protocols such as A2A and MCP. (io.google) Sundar Pichai and other Google executives also used April’s Cloud Next event to frame the company’s enterprise AI push around the “agentic era.” Google’s April AI recap highlighted the Gemini Enterprise Agent Platform and new TPUs as releases aimed at that market. That language matters because I/O often extends platform work from Cloud and developer channels into broader product announcements. That last point is an inference based on Google’s event pattern, not a stated company plan. (cloud.google.com) ### What is WebMCP, and why does it matter here? Chrome Developers introduced WebMCP in early preview on February 10 as a way for websites to expose structured tools to AI agents. Google said the approach is intended to let agents act on sites with greater speed, reliability and precision, and described two APIs: a declarative path for standard HTML form actions and an imperative path for more dynamic JavaScript-based interactions. (blog.google) Google Cloud made a parallel move in December 2025 by announcing official MCP support for Google and Google Cloud services. The company said developers could point agents or MCP clients such as Gemini CLI to managed endpoints instead of stitching together local servers, and said Apigee could expose enterprise APIs as tools for agents. Together, those announcements show Google building both sides of the connection: services that agents can call and websites that can present structured actions. (developer.chrome.com) ### How does Project Astra fit with reports about a more autonomous Gemini? Google DeepMind says Project Astra is “on the way to building a universal AI assistant” and that it is bringing Astra capabilities into Gemini Live, Search and new form factors including glasses. The Astra page says the system can use tools such as Search, Gmail, Calendar and Maps, perform interface control, retain preferences and take actions on a user’s behalf. (cloud.google.com) Those official descriptions overlap with what outside reports have said about “Gemini Spark.” Android Headlines, citing leaks and screenshots, reported on May 15 that Spark could manage Gmail, create reusable “skills,” and handle recurring workflows across Google apps with less step-by-step supervision. Google has not confirmed that name or those features, so they remain unverified ahead of the keynote. (deepmind.google) ### What should developers and site owners watch for on May 19? Google’s own materials point to three concrete areas to watch. First is whether the company names a new Gemini model family or version during the keynote, something its pre-event language leaves open while promising “latest Gemini model updates.” Second is whether Google connects consumer-facing assistant features to the same protocols and tooling it has been rolling out for developers and enterprises. (androidheadlines.com) Third is whether Chrome and Google services get more explicit action interfaces for agents at launch. The next public checkpoint is May 19 at 10:00 a.m. Pacific time, when Google’s keynote begins at I/O 2026. Google has scheduled the developer keynote for 1:30 p.m. the same day, and the company says sessions and on-demand technical content will continue through May 20 on io.google. (io.google) (developers.googleblog.com)