Google is reportedly preparing to take its AI ambitions out of the lab and into the places people already spend their time—starting with Search, and potentially extending all the way to the face. According to CEO Sundar Pichai, the company plans to release smart glasses alongside a new wave of AI “agents” embedded directly within its search experience. The pitch is straightforward but consequential: instead of treating AI as something that merely answers questions, Google wants it to help users complete tasks—moving from “chat” to “do”—while also pairing that capability with a wearable interface that can make the assistance feel more immediate and context-aware.
At the center of this push is Gemini. Pichai’s comments suggest that the next set of features will be powered by a new Gemini model designed to narrow the perceived gap with leading AI competitors such as Anthropic and OpenAI. While Google has already demonstrated impressive capabilities across its products, the competitive landscape has shifted. Users increasingly expect AI to behave like an operator: to plan, to execute, to follow through, and to adapt when circumstances change. That expectation is what makes the “agent” framing so important. It signals that Google is not only improving responses—it is aiming to improve outcomes.
What makes this reported direction different is the combination of two moves that are often discussed separately. Many companies talk about agentic AI in the abstract, but fewer connect it to a mainstream workflow in a way that feels natural. Search is one of the most universal workflows on the internet. People use it to research, compare, decide, and troubleshoot. If agents are integrated into that process, they could become the layer that turns information retrieval into action. Meanwhile, smart glasses represent a different kind of interface shift: rather than asking users to switch contexts—open an app, type a query, read a screen—glasses could bring AI prompts and outputs into the user’s physical environment. Even if the initial hardware is limited, the strategic message is clear: Google wants AI to be present, not just accessible.
The “agent” concept inside Search: from answers to actions
To understand why Google’s reported plan matters, it helps to clarify what “AI agents” typically mean in practice. In the simplest form, an agent is an AI system that can take steps toward a goal. That might involve gathering information, interpreting constraints, generating options, and then performing actions—such as drafting messages, creating checklists, booking services, or guiding a user through a multi-step process. The key difference from a chatbot is that the agent is oriented around completion rather than conversation.
In Search, that could translate into experiences that look less like a list of links and more like a guided workflow. Imagine searching for “best running shoes for flat feet” and not only receiving recommendations, but also getting a plan: how to measure your foot, which models match your gait, what to try first in-store, and how to compare return policies. Now imagine the agent goes further—helping you narrow choices based on budget, availability, and preferences, then generating a short message to a retailer, or compiling a comparison table you can export. The “agent” label implies that Google wants to do more than summarize; it wants to coordinate.
This is where the competitive pressure becomes visible. Anthropic and OpenAI have both pushed the idea that AI should be able to handle complex tasks, not just answer questions. Google’s response, as framed by Pichai, appears to be a bet that Gemini-powered agents can deliver similar end-to-end utility while leveraging Google’s unique strengths: large-scale search infrastructure, deep integration with consumer products, and a long history of building systems that understand web content.
But there’s a nuance here. Search is not a closed environment. It’s a constantly changing ecosystem of websites, services, and user intents. Agents that operate inside Search must therefore be robust against ambiguity and variation. They need to interpret what the user really wants, not just what they typed. They also need to handle uncertainty—when information is incomplete, when sources conflict, or when a task requires user confirmation. The more “agentic” the behavior, the more important it becomes that the system remains transparent about what it is doing and why.
That’s likely part of the reason Google is emphasizing a new Gemini model. A better model alone doesn’t guarantee better agents, but it can improve the underlying reasoning, planning, and instruction-following needed for multi-step tasks. In other words, the model is the engine; the agent is the vehicle. Google’s reported plan suggests it wants both.
Smart glasses: AI that lives in the moment
The second major element—smart glasses—changes the stakes. Wearables are not just another device category; they are a different relationship between AI and context. A phone is a powerful tool, but it often pulls users away from the environment they’re trying to understand. Glasses, by contrast, can keep the user’s attention anchored while still providing information overlays, prompts, or hands-free interaction.
If Google releases smart glasses paired with Gemini-powered capabilities, the company is effectively betting that AI will become more useful when it is less interruptive and more situational. For example, a user might look at a product in a store and ask for comparisons, or receive guidance while assembling something at home. In a travel scenario, glasses could help translate signs, summarize directions, or provide real-time suggestions without requiring constant screen switching.
Even if the initial feature set is modest, the strategic value is in the interface. Search is where users go to ask questions. Glasses could become where users go to act on them. That creates a continuum: discover intent in Search, then execute with the help of a wearable interface that reduces friction.
There is also a broader implication: smart glasses could make AI feel less like a tool you open and more like a companion you consult. That matters because agentic AI depends on frequent, small interactions. The more seamlessly those interactions happen, the more likely users are to trust the system with meaningful tasks.
Of course, wearables introduce their own challenges. Privacy expectations are higher when a device is worn on the body. Context awareness raises questions about what data is captured, how it is processed, and how long it is retained. Google will likely need to address these concerns carefully, especially if the glasses are positioned as an always-available assistant. The success of the product may hinge as much on trust and controls as on raw intelligence.
Why Gemini is central to the strategy
Pichai’s comments tie the entire initiative to Gemini. That’s not surprising—Google’s AI roadmap has been increasingly organized around Gemini as a unifying model family. But the emphasis on a “new Gemini model” suggests that Google believes the next iteration is necessary to deliver the kind of agentic behavior users now expect.
In practical terms, agentic systems require more than fluent language. They need to:
1) interpret instructions reliably,
2) plan multi-step actions,
3) decide when to ask clarifying questions,
4) verify results against available information,
5) handle edge cases gracefully,
6) maintain safety boundaries while still being useful.
A stronger model can improve each of these components. It can also reduce hallucinations and improve grounding—especially important for Search-based agents that rely on external information. If Google wants agents to take actions, it must also ensure that the system does not confidently proceed with incorrect assumptions. That means the model must be better at uncertainty, better at source attribution, and better at knowing when to stop and ask the user.
Google’s advantage is that it can ground responses in the web and in its own knowledge systems. The challenge is that grounding is not the same as correctness. Agents must not only cite information—they must use it appropriately to complete tasks. That is a higher bar than answering a question.
So the reported plan reads like a bet that Gemini is ready for that higher bar, at least in targeted scenarios.
Closing the gap: what “gap” really means
When tech leaders talk about closing the gap with OpenAI and Anthropic, they often refer to more than model quality. They refer to product experience: how quickly users can get value, how well the system handles complex requests, and how naturally it fits into daily workflows.
OpenAI and Anthropic have benefited from a strong narrative around reasoning and agent-like capabilities. Google, meanwhile, has historically excelled at search relevance and large-scale systems, but the AI interface has sometimes felt like an overlay rather than a replacement. The reported move suggests Google wants to make AI feel native to Search—less like a feature and more like the new default way to interact with information.
If Google succeeds, the “gap” could narrow in several dimensions:
– Task completion: agents that finish work rather than just respond.
– Integration: AI that connects to tools and services users already rely on.
– Responsiveness: faster, more context-aware interactions.
– Trust: clearer explanations and safer behavior.
– Multimodality: potentially richer inputs and outputs, especially if glasses add a new channel for context.
The smart glasses piece could also be part of the “gap” narrative. Competitors have focused heavily on chat interfaces and developer ecosystems. Google’s wearable approach could differentiate the experience by making AI more ambient and less screen-bound.
What this could mean for the future of Search
Search has been evolving for years—from blue links to rich answers, from autocomplete to knowledge panels, from desktop to mobile-first experiences. The next evolution, if the reports are accurate, is that Search becomes a task engine.
That would be a major shift in how people think about searching. Instead of asking, “What is the best option?” users might ask, “Help me do this,” and the system would handle the intermediate steps. This is not just a UX change; it changes the economics of attention. If agents reduce the need to click through to multiple sites, traffic patterns could shift. Publishers and businesses that rely on search referrals may need to adapt to a world where AI summarizes and acts without sending users to the same extent as before.
Google will likely try to balance this by ensuring that agents remain