Google I/O on May 14th was two hours of AI announcements. Project Astra, Gemini 1.5 Flash, AI Overviews in Search, Veo for video generation, Gemini in Workspace. Google's message was impossible to miss: every product is now an AI product. The question is whether any of it sticks.
Project Astra and the universal AI agent vision
Project Astra was the demo that got the most attention. A model that can see through a phone camera, understand what it is looking at in real time, remember context across a conversation, and respond to voice naturally. The demo showed someone using it to identify code on a screen, find where they left their glasses, and debug a circuit board. It looked like the universal AI assistant concept that research labs have talked about for years actually working.
Google said Astra capabilities will start showing up in Google products later this year. The underlying technology is Gemini 1.5 Pro with a continuous context window and real-time multimodal processing. The infrastructure challenge of running this at Google Search scale is non-trivial, which is probably why they are rolling it into products gradually rather than launching it standalone.
Gemini 1.5 Flash
Flash is the more immediately useful announcement for developers. It is a smaller, faster, cheaper version of Gemini 1.5 Pro that retains the long context window. For applications where you need the context length but not the full reasoning capability of Pro, Flash is the cost-effective option. It launched in the API at the event with competitive pricing.
AI Overviews in Search
AI Overviews, previously called Search Generative Experience, is now the default for many queries in the US. Google is summarising search results with AI before the traditional link results appear. This is the change with the largest real-world impact: it affects how people find information and how publishers get traffic. The early rollout has had some high-profile errors that attracted attention, but the feature is shipping broadly regardless.
For anyone building content or products that depend on organic search visibility, AI Overviews changes the calculus. The click-through on traditional results goes down when a summary answers the query. That is a structural shift, not a temporary experiment.