Google launches background AI-agents in search — for now only for Ultra subscribers

Google has begun real deployment of search AI agents announced at the May Google I/O conference. For now, access to the feature is available only to subscribers of the most expensive plan — Google AI Ultra.

What is a background search agent

At Google I/O in May 2025, the company unveiled a fundamentally new approach to search — agents capable of working autonomously, without constant user control. These agents are built on top of the Gemini model and can independently break down complex queries into subtasks, access multiple web sources simultaneously, synthesize findings, and return a structured answer. The key idea: the user sets a goal, not a query.

Previously, for example, to compare mobile operator rates, you'd need to open a dozen tabs, read each page, and manually consolidate the data. The agent does all this in the background — while the person is busy with other things — and returns a ready-made digest. Background mode means you don't need to wait with an open tab.

Just formulate your request, minimize the browser — and get a notification when the agent is done.

Who got access first

The feature is currently available only to Google AI Ultra subscribers. This is Google's flagship pricing plan, aimed at users who actively work with the company's AI tools.

• Google AI Ultra — the most expensive plan, includes priority access to new features
• Ultra subscribers get experimental features before wide release
• Google hasn't announced specific timelines for other plans
• By analogy with past launches, expansion typically takes several months

This approach allows the company to gather real user feedback in a controlled environment. Launching an experimental feature to hundreds of millions of users at once would be risky both in terms of reputation and infrastructure load.

Why this changes the search model

All classical web search is reactive: the user enters a query, gets a list of links, reads them, synthesizes them. The AI agent flips the scheme: takes a task, breaks it into pieces, goes through several iterations, and delivers a ready result. For regular users, this means that multi-step tasks — for example, 'find the best laptop options for $1,500, compare specs, and check current prices' — are now solved with one query, not a half-hour session with a dozen open tabs. For business, the potential is even wider: regular competitor monitoring, automatic industry news digests, mention tracking by specified parameters.

'Agents do for you what previously required multiple search sessions,' — describes the concept in

Google's official blog.

Competition is accelerating

Google is deploying agents not in isolation. OpenAI already offers Deep Research in ChatGPT — a multi-step research function for paid subscribers. Perplexity is building its own agent ecosystem. Microsoft has integrated agent capabilities into Copilot for enterprise users. The race for next-generation intelligent search is unfolding across the market simultaneously. Google has an obvious advantage — the largest search index and global infrastructure. But while Ultra subscribers test agents in background mode, competitors are already working with a broader audience.

What this means

Google is moving search from a tool to an assistant. The transition from the May demo to real deployment is an important signal: the technology is ready for real users, not just a conference stage. How quickly agents become standard for everyone will be shown over the next few months.