Google Search adds AI agents and upgrades to Gemini 3.5 Flash as default

Google Search just got its most significant overhaul in years. At I/O 2026, Google announced that Gemini 3.5 Flash is now the default model powering AI Mode globally, introduced a new class of persistent AI agents built directly into Search, and redesigned the Search box itself for the first time in over 25 years. These changes are live now or rolling out this summer, and they affect every Google Search user.

Here is what actually changed, and what it means in practice.

Gemini 3.5 Flash is now the engine behind AI Mode

AI Mode, Google’s conversational search experience, has passed one billion monthly users. Queries in AI Mode have been more than doubling every quarter since launch, and overall Search query volume hit an all-time high last quarter. That growth prompted Google to upgrade the underlying model.

Gemini 3.5 Flash replaces the previous default, and the jump is meaningful. It is Google’s strongest agentic and coding model to date, outperforming Gemini 3.1 Pro on benchmarks including Terminal-Bench 2.1 (76.2%) and MCP Atlas (83.6%). More practically, it is four times faster than comparable frontier models in terms of output tokens per second, which matters a lot when Search is generating dynamic, structured responses in real time.

For everyday users, this means AI Mode responses should be noticeably faster and more capable, especially for complex, multi-step questions. You do not need to do anything. The upgrade is automatic and global.

Information agents: Search that works while you are not there

The headline feature is what Google calls information agents, and this is where Search starts behaving differently from anything it has done before.

Rather than answering a question once, an information agent monitors the web continuously on your behalf. It watches blogs, news sites, social posts, financial data, shopping listings, sports scores, and anything else relevant to your query. When something meaningful changes, it sends you a synthesised update rather than a raw data dump. You can run multiple agents simultaneously, each tracking different topics or tasks.

Setting one up is simple: add “keep me updated” to any search query. Your active agents appear in a side panel in AI Mode.

To start, this feature is for Google AI Pro and Ultra subscribers and launches this summer. If you are researching a home purchase, following a developing news story, tracking a product price, or monitoring a competitor, this is a meaningful shift in what a search engine can do for you. The work moves from you checking back repeatedly to Search telling you when there is something worth knowing.

Google has also redesigned the Search input box for the first time in over 25 years. The new box expands dynamically as you type, supports AI-powered suggestions that go well beyond standard autocomplete, and accepts text, images, files, videos, and Chrome tabs as inputs.

This is not cosmetic. The redesigned box is intended to help you articulate what you actually need, particularly for complex queries where a short keyword phrase would previously have fallen short.

Generative UI and custom dashboards

When you ask a complex question in AI Mode, Search can now build a custom visual response on the fly. Rather than a standard text answer, it might assemble interactive graphs, tables, simulations, or structured layouts suited to your specific question. Google is calling this Generative UI, and it is powered by a combination of Gemini 3.5 Flash and Google’s internal agent-first development platform, Antigravity.

For ongoing tasks like planning a wedding or managing a home move, Search can go further and generate a persistent dashboard or tracker you can return to over time. These custom dashboards are coming in the next few months, initially for AI Pro and Ultra users in the US.

Generative UI for standard queries will be available free to all Search users this summer.

Agentic booking, including Google calling businesses for you

Agentic booking in Search is expanding this summer across the US. You can describe what you need in specific terms (“a private karaoke room for six on a Friday that serves food late”) and Search will pull together live pricing, availability, and booking links.

For certain categories including home repair, beauty, and pet care, you can ask Google to phone the business on your behalf. This is a significant step from clicking links to having tasks completed without you picking up the phone. It rolls out to all US users this summer.

Seamless follow-up from AI Overviews

One smaller but useful change is already live worldwide: you can now ask a follow-up question directly from an AI Overview and continue into a full conversational thread in AI Mode. Context carries over, and supporting links become more relevant as the conversation deepens. This works across desktop and mobile, no subscription required.

The practical summary

Most of these changes touch you in the near term regardless of whether you pay for a Google subscription:

Right now, free: Gemini 3.5 Flash is the default in AI Mode globally. Faster, smarter responses across the board.
This summer, free: Generative UI for dynamic visual responses in Search. Agentic booking in the US.
This summer, AI Pro and Ultra: Information agents for continuous web monitoring. Custom dashboards and trackers.

Google is clearly moving Search from a tool you query to a service that does ongoing work for you. Information agents in particular represent a different model of interaction, one where you set an intent once and get notified rather than checking back manually. Whether that lands as useful or intrusive will probably depend on how well the relevance filtering works in practice.

For developers, Gemini 3.5 Flash is available now via the Gemini API, Google AI Studio, and Antigravity, at $1.50 per million input tokens and $9.00 per million output tokens.