Google AI Mode rolls out search agents that track information for you

At a glance:

• Google launches information agents in AI Mode for AI Ultra subscribers at I/O 2026
• Agents monitor web, finance, shopping, and sports 24/7 and deliver synthesized updates instantly
• Feature expands to AI Pro this summer with trigger phrases like "keep me updated on"

What happened at I/O 2026

Google used its I/O 2026 developer conference to introduce the concept of "Search agents," beginning with information agents that now roll out inside AI Mode for AI Ultra subscribers. The company describes these agents as background workers that operate around the clock and "intelligently reason across information to find exactly what you need at exactly the right moment." Unlike traditional alerts or scheduled summaries, the agents continuously scan blogs, news sites, social posts, and other web content alongside real-time finance, shopping, and sports data to detect changes tied to a user's specific question.

The initial rollout targets AI Ultra, Google's highest consumer AI tier priced at $99.99 or $199.99 per month depending on the plan. Subscribers can activate an agent by including phrases such as "keep me updated on" or "alert me when" in their AI Mode prompt. Google says the capability is available across all AI Mode languages and markets from day one, signaling a global launch rather than a limited regional test.

How information agents work

In practice, an information agent produces a "synthesized update, with the ability to take action" whenever it discovers relevant new material. Google's keynote example asked the agent to monitor for sneaker collaborations or signature drops involving a user's favorite athletes; the agent would then surface a concise briefing the moment such announcements appear online. Another illustration involved listing apartment requirements — price range, neighborhood, amenities — and receiving a notification only when new listings match every criterion.

The agent's reasoning layer is designed to filter noise and avoid the flood of irrelevant matches that plague keyword alerts. By combining large-language-model comprehension with live search indexing, the system can interpret nuanced constraints such as "only notify me if the collaboration includes a limited-edition colorway" or "exclude listings without in-unit laundry." This moves the product closer to a personal research assistant than a simple feed filter.

Pricing and availability

AI Ultra remains the exclusive home for search agents at launch, with two price points reflecting different compute and storage allowances. The $99.99 tier targets individual power users, while the $199.99 tier adds higher rate limits and priority access to new agent types as they debut. Google has committed to bringing the feature to AI Pro — its mid-range plan — "this summer," though no exact date or pricing adjustment was disclosed.

The phased rollout mirrors Google's strategy with previous AI Mode capabilities, where Ultra serves as a proving ground for resource-intensive features before they are optimized for broader tiers. Company representatives emphasized that the underlying infrastructure scales horizontally, so the summer expansion should not require architectural changes.

Comparison with existing Gemini features

On paper, the main advantage over the Gemini app's Scheduled actions — which run up to once per day — or Gemini Spark's 15-minute polling interval is the immediacy with which Search agents can deliver updates. Scheduled actions are essentially cron jobs that execute a prompt at a fixed time, while Spark's cadence still leaves a window of up to 15 minutes between checks. Information agents, by contrast, leverage Google's continuous indexing pipeline to push findings the moment they are crawled and ranked.

This architectural difference matters for time-sensitive domains such as flash sales, breaking sports trades, or regulatory filings that move markets. However, Google has not published latency benchmarks, and the real-world gap will depend on crawl frequency for specific source types. The company also notes that agents can "take action," hinting at future integration with shopping carts, calendar booking, or form submission — capabilities absent from the current scheduled-action framework.

What to watch next

The summer expansion to AI Pro will be the first real test of whether the agent architecture can serve a much larger user base without degrading freshness or increasing hallucination rates. Analysts will also watch for the arrival of additional agent types beyond information monitoring; Google's I/O slides referenced "task agents" and "creative agents" as future categories. Meanwhile, competitors such as OpenAI's Operator and Anthropic's computer-use demos suggest the industry is converging on persistent, goal-oriented agents that blend search, reasoning, and tool use.

For now, AI Ultra subscribers gain a differentiated feature that leverages Google's unique combination of search index breadth and Gemini model depth. The pricing premium positions agents as a flagship capability rather than a commodity add-on, and the summer timeline for AI Pro will indicate how quickly Google can democratize the technology without compromising its core value proposition.