Has Google Reclaimed the Throne of the AI Race with Gemini 3? Decoding the “AI That Builds UIs” and Agent Strategy

Google has announced its latest large language model, “Gemini 3 (hereinafter Gemini 3),” and the Silicon Valley AI scene is buzzing.
On the Hard Fork podcast, Google DeepMind CEO Demis Hassabis and Gemini team VP Josh Woodward appeared to discuss “why Gemini 3 pushes Google back into the lead pack in the AI race.” In this article, we will organize that content and explain the technical aspects in as accessible a way as possible.

1. Why is “Gemini 3” getting a special episode?

Podcast hosts Casey and Kevin preface the episode by saying they usually “don’t do special episodes for every new model,” but emphasize that this time is an exception. There are three main reasons for this.

1. Direct interview with the core of Google
   The guests were two key figures at the heart of Google AI:
   ・Google DeepMind CEO Demis Hassabis
   ・Gemini VP Josh Woodward
   It was an opportunity to hear directly about the model's philosophy and strategy from the people holding the reins of both “research” and “product deployment.”
   

2. A model that competitors are clearly “wary” of
   It is said that “whispers” are spreading among people at other AI labs that “Gemini 3 is at a level that is quite troublesome for business.” Because there is a sense that Google, which was seen as the “chaser” with Bard and early Gemini, has returned to being an entity asked, “Has it returned to the top of the AI leaderboard?”

3. A “step-change” performance jump from the previous generation
   A prime example cited was the difficult benchmark called “Humanity’s Last Exam,” which is at the graduate to doctoral level.

   • Gemini 2.5 Pro: 21.6%

   • Gemini 3 Pro: 37.5%

The score has improved significantly. Hassabis positions this as “one of the fastest advancements in the industry in the last few years.”

2. What is new: “Interface generation” that goes beyond chat and agentification

2-1. A model that “builds and returns a UI” instead of just an “answer”

Woodward emphasized that Gemini 3 generates the “interface itself” rather than just a text response.

• Write a question in a form
  → Automatically generates an “interactive tutorial UI” instead of a text response

• Consult about a mortgage
  → Builds a “mortgage calculator tool” on the spot, not just a text explanation

The host introduced a demo where a learning app-like screen containing images and interactive elements was automatically generated for a user wanting to learn about Vincent van Gogh.
It can be said that it is moving closer to the positioning of an “AI that builds apps on the fly” rather than the “chatbots” of the past.

2-2. “Gemini Agent” that dives into your mailbox

Another highlight is the “Gemini Agent” that Google is testing. This reads the user's Gmail (with permission, of course):

• Understand the content and suggest reply drafts

• Bundle and organize related emails

• Take over the “organization tasks” of the inbox itself

It performs things like these.

The host said it is a “feature I want to try as soon as it comes out,” but at this point, the official launch date for Docs or Gmail add-ons is undecided. It will first be rolled out via the Gemini app and the AI mode in Search.

3. Gemini as a Learning Tool: A Strategy to Capture the Entire Student Demographic

Google also announced a bold initiative this time: offering the paid version of Gemini for free for one year to all university students in the United States as well.

During the briefing, multiple representatives repeatedly

However, the host pointed out, half-sarcastically, "Isn't this essentially a 'homework-doing tool'?" highlighting the structure of Google's attempt to capture the next generation of users in their entirety.

4. Has Google Returned to the "Front" of the AI Race?

4-1. Hassabis's Self-Assessment: "Speed of Progress" Over the Race

When asked the direct question, "Is Google currently at the front of the AI race?" Hassabis responded as follows.

• The competition is "probably the most intense in history."

• What matters is the speed of progress, not where we are right now.

• In research, there is a sense of pride in "having always been at the forefront."

He then emphasized the strategy of using Gemini as the "engine room for all of Google,"

• Search

• Gmail / Workspace

• YouTube

• Android / Maps

and integrating it into the back end of existing products. The difference from AI-focused startups is that they "already have services used by billions of people." The host half-jokingly

4-2. The Outlook of "5 to 10 Years" to AGI Remains Unchanged

Hassabis has previously stated that it would take "5 to 10 years to reach AGI (Artificial General Intelligence)," and looking at Gemini 3, he has not changed this timeline.

He stated this, suggesting that expansion into "physical intelligence" by combining projects like Simmer and Genie will be the key.

5. Safety, Risks, and the Bubble Theory: Caution Due to Powerful Tools

While Gemini 3 has been significantly enhanced in terms of performance, safety and misuse risks have naturally increased as well.

• Improved tool-calling and function-calling capabilities

• While useful for code generation and automation, this can also lead to misuse such as cyberattacks

Hassabis emphasized that "this is the most thoroughly verified model, developed in collaboration with internal testing as well as external safety evaluation organizations and testers," while also showing a cautious stance by noting, "The more capable it becomes, the more we must keep a close eye on cyber risks."

Furthermore, when asked if this is an "AI bubble," he replied:

• There are clearly bubble-like aspects, such as areas where billions of yen are moving in seed rounds

• However, there are also many "green fields" that could become massive industries in the long term, such as robotics, gaming, drug discovery, and cloud/TPU,
  he said, adding,

Conclusion: Is Gemini 3 the Starting Gun for the Full-Scale Race Beyond Chat?

What becomes clear through this podcast is that Gemini 3 is not just a "high-accuracy chatbot," but is evolving into:

• A generative interface engine that reconfigures the UI for each user

• An agent that dives into emails and documents to perform tasks on behalf of the user

• An AI that acts as a "super tool" spanning learning, creation, and work efficiency

The picture is one of evolution.

Hassabis still sees the road to AGI as being "5 to 10 years" away, but in that process,

• Google, which is redesigning its existing products to be AI-first,

• and AI-specialized startups competing solely on models will find themselves in an increasingly distinct "battle on different playing fields."

Gemini 3 can be called the first round's gong—a move that makes one feel Google has come to 'take back the crown' once again.

Recommended Articles

Next Big Wave (Growth Stocks, Seeds of Ideas, and Trend Deep Dives)