Google placed artificial intelligence at the center of its next generation of products this week, unveiling a conversational assistant designed to follow users from search to a new living-room speaker and into messaging. The strategy, introduced at the company’s annual developer conference, seeks to turn Google’s vast index and machine-learning systems into a service that can understand context, complete transactions and respond in ordinary language.

Chief Executive Sundar Pichai described the shift as a move from a mobile-first world to an “AI-first” one. In the company’s official account of the May 18 keynote, Google said its Assistant would build on two decades of search while learning a user’s preferences and surroundings. It will first appear in Google Home, a voice-controlled speaker due later this year, and Allo, a new messaging application for Android and Apple devices.

An assistant that follows the conversation

The distinction Google is drawing between its existing voice search and the new Assistant is continuity. Rather than treating every question as an isolated request, the system is intended to retain conversational context. A user might ask what films are playing, narrow the answer to child-friendly titles, and then buy tickets without restating the subject. TIME’s demonstration account noted that the assistant could answer a follow-up question about a film director without requiring the person’s name to be repeated.

That capability rests on Google’s advances in speech recognition, natural-language processing and image analysis. Pichai said 20 percent of searches in Google’s mobile application and on Android devices are now made by voice, a sign that speaking to computers is moving beyond a novelty. The company’s aim is broader than dictation: it wants the assistant to anticipate what information is useful and present choices that lead to an action.

The challenge is that conversational systems can hide their own capabilities. A graphical application displays buttons and menus; a user facing a microphone may not know what to ask. Wired’s examination of the interface described Google’s use of suggested replies and prompts to teach people what the assistant can do while keeping the exchange moving.

Google Home carries search into the living room

Google Home turns that software into hardware. The compact, Wi-Fi-connected speaker will answer questions, play music, manage everyday tasks and control compatible household devices. Multiple units can be grouped, allowing music to be sent to one room or throughout a house. The product places Google in direct competition with Amazon’s Echo and its Alexa service, which already established a market for voice-controlled home computing.

Unlike a phone, Home is meant to be shared and available without being picked up. That convenience also raises difficult questions about identity and privacy: a household device must distinguish users, protect calendars and messages, and decide how much of a conversation should be retained. Google offered no price this week and said the product would arrive later in the year. A contemporaneous conference review by Urban Airship emphasized that the speaker’s importance lies in carrying the same Assistant into music, search, tasks and connected-home controls rather than operating as a separate appliance.

Allo and Duo put machine learning inside communication

Allo, scheduled for release this summer, inserts the Assistant directly into personal conversations. Users can ask it for restaurant suggestions, directions or sports scores without leaving a chat. Its “Smart Reply” feature proposes short responses based on the content and tone of a message and can recognize objects in photographs. Google is betting that a messaging service becomes more valuable when the software can contribute information and actions inside the discussion.

The service will be tied to a telephone number and released on both Android and Apple’s iOS, a cross-platform approach intended to reduce the barrier to adoption. TIME’s account of the introduction detailed how Allo can recommend restaurants and begin an OpenTable reservation while a group is deciding where to eat. Duo, a companion one-to-one video-calling application, will also use phone numbers and will preview a caller’s live video before the recipient answers.

Encryption is a point of tension. Duo calls will be protected end to end, but Allo will use that protection only in an optional Incognito mode. Ordinary Allo conversations will be available to Google’s systems so the Assistant and Smart Reply can function. Wired reported that Incognito mode uses the Signal protocol and can set messages to expire. The design comes as Apple and the Federal Bureau of Investigation have fought publicly over access to encrypted devices and as WhatsApp has enabled end-to-end protection by default for more than a billion users.

Android expands into virtual reality and instant apps

The Assistant was the keynote’s organizing idea, but Google also broadened Android. Daydream, a virtual-reality platform built into the coming Android N release, will define performance standards for phones and provide a reference design for headsets and handheld controllers. Google said manufacturers including Samsung, HTC, LG, Xiaomi and Huawei were working on compatible phones. The Verge’s roundup of the announcements described Daydream as an Android-based ecosystem for mobile virtual reality, with adapted versions of YouTube, Street View, Google Photos and the Play store.

Android Instant Apps address a different source of friction: the need to install a full application before completing a small task. A link could download only the components needed to view a video, make a purchase or pay for parking, then disappear when the task is finished. Google said the feature would work on versions of Android dating to Jelly Bean and would begin reaching users later this year. Ars Technica’s technical review explained that developers would modularize applications into small pieces delivered on demand, narrowing the gap between native apps and the web.

Android N will also adopt seamless system updates modeled on Chrome OS, installing a new version in the background and switching to it at the next restart. Ars Technica reported that the change is designed to eliminate the long “installing system update” pause, though it will initially depend on new devices built with two system partitions.

A platform contest moves beyond the screen

Google’s announcements amount to a contest over the layer through which people reach information and services. Amazon has established an early position in the home; Facebook dominates mobile messaging; Apple controls a tightly integrated combination of hardware, operating system and services. Google’s answer is to make its strongest asset—search informed by machine learning—available across devices and conversations.

The strategy carries execution risks. Home has no announced price, Allo must persuade users to add another messaging application, and the Assistant’s usefulness will depend on outside developers connecting their services. Privacy choices may also determine whether people trust an always-listening speaker or a chat service that analyzes messages.

Yet the breadth of the rollout is consequential. Google is no longer presenting artificial intelligence as a feature attached to search. It is building a common interface for phones, homes, messages, watches and virtual environments. If developers and consumers adopt it, the company’s familiar search box may become only one doorway into Google’s services—and perhaps no longer the most important one.