Google expands consumer AI agents

2026-05-20

Rollout of consumer features begins

US Google on the 19th announced it will launch features that let artificial intelligence (AI) continuously take over tasks such as search, email and reservations. Using its strength in services with more than 3 billion users, the company is extending AI agents, now spreading in enterprise use, to consumers as well.

At its annual developer event Google I/O held the same day at its California headquarters, Google unveiled new services that include multiple agent functions. Chief Executive Officer Sundar Pichai said in the keynote that the company will focus on safely bringing the power of agents to consumers and making them available to everyone.

Spark handles complex tasks

The new consumer feature Gemini Spark is available in the conversational AI Gemini and other services. Even for tasks that span multiple apps such as email, calendars and document creation, AI can grasp intent, gather needed information and carry out the work. It also supports voice commands.

For example, it can extract important schedules and deadlines from emails sent by a child's school and share them daily with a spouse, or check RSVP replies for a party, create a packing list and draft a follow-up email prompting responses.

Because the instructed agent runs in the cloud, it keeps working even when a PC or smartphone is powered off, such as during sleep. Google also plans integration with services outside its own ecosystem and with its Chrome browser, allowing it to handle tasks across web pages.

Available to paid users next week

Spark will be offered from next week to US paid subscribers on plans starting at $100 a month, or about 15,900 yen. Google also indicated it plans to widen access, saying it will eventually open the feature to all users, including free accounts.

In search, the company will launch an agent feature called always-on search this summer. Users can specify items such as vacant properties that meet certain conditions or merchandise from a favorite athlete, and AI will repeatedly search at regular intervals and notify them when matching results appear.

The company will also begin rolling out an agent feature for e-commerce in the United States. When an item is added to the cart, the system can continue comparing price histories and inventory status across multiple sites, and it will also allow checkout of items from separate e-commerce sites in one place.

Model updates continue

Among AI models, Google announced Gemini Omni, designed for high-quality video generation and editing. It is said to understand the physical laws of the real world and produce more realistic footage, and will be offered to paying users.

The company also unveiled Gemini 3.5 Flash, its latest core model in about three months. It reportedly outperforms previous high-end models on coding and agent-function benchmark scores, while its output time is one-quarter that of rival products. It will be made available to all users through Gemini and other services, and also built into Google services such as search and agents.

In AI development, OpenAI, Anthropic and Google continue to battle for leadership. While Anthropic's private Mythos announced in April drew attention, Google did not present an AI that matched the top-performing models in this round of announcements. On the US stock market on the 19th, shares of parent Alphabet fell 2% from the previous close.

Enjoyed this article? Share it with your network!