Google’s flagship AI model is no longer being built to answer questions better. In a published interview with Google DeepMind SVP and Chief AI Architect Koray Kavukcuoglu, Kavukcuoglu stated that Gemini is increasingly seen as a Gemini agent rather than a chatbot or language model. The transition has direct implications for how AI-powered search surfaces, selects, and acts on brand information.
What Does It Mean for Gemini to Become an AI Agent?
An AI agent (a system that takes multi-step actions on your behalf, not just generates a text response) is fundamentally different from a chatbot. Per the Search Engine Journal interview, Kavukcuoglu explained that the goal is not really about creating a model that answers questions better, the focus now is to create something that can take actions on behalf of and alongside a human. This repositions Gemini from a retrieval-and-response tool to an active participant in tasks that span multiple steps and tools.
Why Did Coding Become the Gateway to Agentic AI?
According to Kavukcuoglu in the SEJ-reported interview, working on software engineering problems forced the team to teach Gemini how to use tools, call functions, and execute multi-step workflows. “Basically turn this whole thing into an agent, from a model to an agent,” he said. He noted that improvements seen in Gemini 3.6 and 3.7 trace back to parallel research tracks running over a year earlier, converging into what the team shipped.
Has the Underlying AI Technology Changed as Dramatically as the Results Suggest?
Kavukcuoglu’s answer, as reported by Search Engine Journal, is largely no. He said the optimization techniques powering Gemini rely on similar underlying principles as they always have. What has changed is the environment: models now need to infer user intent (understanding what someone actually wants, not just what they typed), handle ambiguity, and collaborate with humans across extended, multi-step tasks. Revolutionary outputs do not necessarily require revolutionary methods.
What Is Google DeepMind Still Figuring Out?
Kavukcuoglu acknowledged ongoing learning. The SEJ interview reports he said Gemini 3.5 specifically taught the team how people actually work with agents in practice, and that confidence has grown from there. “The frontier will always shift,” he noted, and the cadence of which AI lab produces the most capable model will continue to change. When asked about architectural innovation happening inside Google DeepMind, he declined to elaborate, suggesting further structural changes may be in development. On what single improvement he would prioritize, his answer was direct: “I think if I had a magic wand, I would just make them more intelligent.”
What This Means for AI-Search Visibility
Here is the plain-language version of why this matters for brands. When someone asks Gemini a question today, the model retrieves information and writes a response. An agentic model does more: it might search, compare options, navigate to your website, and recommend a specific vendor with a reason. Your brand’s visibility shifts from “does the AI mention me?” to “does the AI choose me when completing a task for a user?” Those are related but distinct challenges.
The following is Hingewise’s read on the signals in this shift, not a statement from the source interview:
- Being cited is not the same as being chosen. Current AI-search optimization focuses on getting models to surface and quote your content. An agentic model may consider multiple sources but ultimately select one for a task. The criteria for that selection are not yet fully transparent, but specificity and authority likely matter more than volume of mentions.
- Structured, task-ready content may carry more weight. Agents need to understand quickly what a brand does, who it serves, and what makes it distinct. Content built around clear differentiators, specific service scope, and concrete evidence is easier for an agent to act on than generic brand narrative.
- Google’s full product ecosystem is now in scope. Kavukcuoglu noted, per SEJ, that all of Google’s products including Gmail, Sheets, and Maps are designed to help people accomplish tasks. If Gemini becomes the connective layer across those products, brands with consistent, accurate presence across Google’s data surfaces may have a structural advantage in agent-driven recommendations.
- Task-completion AI and answer-based AI require different optimization thinking. Most current AEO (Answer Engine Optimization, strategies to get AI models to quote your content) is built for models that read and retrieve. Agentic AI reads, retrieves, and acts. These overlap but are not identical, and it is worth assessing which phase your current content approach is built for.
Hingewise’s assessment: the AI industry has spent roughly three years optimizing for models that answer. The next phase is optimizing for models that complete tasks on a user’s behalf. The two require related but distinct approaches. What counts as a “good answer” for a retrieval model and what counts as a “reliable recommendation” for an agentic model are not the same question.
Quick Checklist: What to Review Before Agentic AI Scales Further
- Does your core offering appear clearly in the first paragraph of your key landing pages, not buried deep in body copy?
- Is your structured data (schema markup, a machine-readable annotation layer added to your website’s code) current and accurate for services, products, and locations?
- Does your content address multi-step decision queries (“how do I choose a vendor for X”) not only single-turn informational queries (“what is X”)?
- Is your brand described consistently across Google Business Profile, your own website, and third-party sources?
- Do your key pages give an AI agent a specific, verifiable reason to recommend you over a competitor?
- Are your pages technically accessible to AI crawlers (check your robots.txt for unintended access blocks)?
The clearest signal from this interview is that Google is not building a better search box. It is building a system designed to complete tasks for users. How that changes the AI-search landscape for brands, and at what pace, is the question worth watching closely as Gemini’s agentic capabilities continue to ship.
Source: Search Engine Journal, reporting on a published Google DeepMind interview with SVP Koray Kavukcuoglu.
Sources
This article was drafted with AI assistance from the sources cited above and checked against them before publishing. The "What this means for AI-search visibility" section is Hingewise's editorial analysis, built on the frameworks and positions we use in our GEO work. How we work: Editorial policy.