The search box used to be a door.
You typed a thing, Google gave you links, and then you walked through one of them.
Google is now trying to turn that door into a desk.
At I/O 2026, the company announced what it calls the biggest upgrade to the Search box in more than 25 years. The new version expands for longer conversational questions, accepts text, images, files, videos, and Chrome tabs as inputs, and routes users into AI Mode with Gemini 3.5 Flash as the default model.
That is not just a nicer query box.
It is a different contract.
Search starts doing the work
The old bargain was simple: Google found pages, you did the reading.
The new bargain is closer to delegation. Google says people will be able to create and manage information agents directly inside Search. These agents will run in the background, monitor the web and live data, then send synthesized updates when a condition is met.
Apartment listings, product drops, sports updates, market moves, and local availability all fit the same pattern.
You no longer search once.
You describe an ongoing need and let the system watch for it.
Google is also adding more agentic booking features. For some local services, Search can gather pricing and availability, then give direct links to finish the booking. In some categories in the United States, Google says users will be able to ask it to call businesses on their behalf this summer.
This is search as task routing.
Mini apps inside Search
The stranger part is generative UI.
Google says Search will be able to build custom layouts, tools, simulations, dashboards, and trackers on the fly. Ask about a complex topic and Search might assemble an interactive visual instead of a normal result page. Ask about an ongoing task and it might create a little dashboard you return to later.
That matters because it attacks the website from two sides.
First, Search summarizes the web before the user clicks.
Then it starts creating the interface the user would have gone elsewhere to use.
For publishers and product teams, this is the uncomfortable part. Google is not only competing for the answer. It is competing for the workflow after the answer.
The real signal
Google is trying to defend Search by making it less like Search.
That sounds weird, but it is rational. If ChatGPT, Claude, Perplexity, and agentic browsers train users to ask bigger questions and expect completed work, a ranked page of links starts to feel thin.
Google has the web index, live commercial data, Maps, Gmail, Photos, Chrome, Android, and the distribution to make agents feel native instead of bolted on.
That is the advantage.
The risk is the same one it has always been with AI Search: if Google becomes the place where the answer, the interface, and the task all happen, fewer people will touch the original web.
Search used to send you somewhere.
Now Search wants to keep the job.
Sources: Google, TechCrunch