NextFin News - This article reports on a special interview conducted for the New York Times technology podcast Hard Fork on 2025-11-18, in which Demis Hassabis, chief executive of Google DeepMind, and Josh Woodward, vice president of Google Labs and the Gemini team, spoke with hosts Kevin Roose and Casey Newton about Google’s newly launched Gemini 3 model and the company’s product strategy around it. (blog.google)
The conversation was recorded just ahead of Gemini 3’s public rollout and accompanied an early product briefing from Google. The guests described the model’s technical strengths, practical applications and safety work, and placed Gemini 3 within Google’s broader plan to integrate advanced AI across Search, Workspace, the Gemini app and developer tools. (techcrunch.com)
What Gemini 3 does: reasoning, generative interfaces and coding
Both Hassabis and Woodward emphasized three areas where Gemini 3 represents visible progress. Woodward highlighted reasoning improvements, saying the model now "really excel[s] on reasoning and being able to think many steps at the same time," and noted that it is much better at maintaining its train of thought than earlier versions. Demis Hassabis underscored the model’s ability to produce new kinds of outputs: "all kinds of new generative interfaces... our best model yet at being able to create new types of interfaces."
this is our best model yet at being able to create new types of interfaces— Demis Hassabis
On coding, both executives said the team invested heavily in developer-oriented capabilities. Woodward and Hassabis described improved "vibe coding" and front-end coding support and pointed to a new agentic coding product (referred to in the briefing as Google Antigravity) that showcases how Gemini 3 can be used inside a multi-pane development environment.
Benchmarks and product-readiness
The guests presented benchmark gains as evidence of technical progress while cautioning that benchmarks are proxies for real user value. Woodward noted large jumps over the prior Gemini 2.5 Pro on difficult tests, and Hassabis said the model’s performance is "dead on track" with their expectations. Both said the ultimate test is translating benchmark improvements into product experiences that users notice and value.
Generative UIs and new user experiences
Hassabis and Woodward described demonstration examples in which Gemini 3 composes interactive, task-focused interfaces on demand. They detailed examples shown at Google briefings — an interactive Vincent Van Gogh tutorial and a mortgage calculator that generated custom controls and visual elements — and framed these as a new class of generative user interface rather than simple text output.
it just sort of like coded up an interactive tutorial that had all sorts of images and interactive elements— description of a Gemini 3 example
Rollout plan and product integration
Both executives explained that Gemini 3 would be rolled out first in the Gemini app and in the AI mode tab of Google Search, and would be made available to developers in other products. Hassabis stressed Google’s ability to serve large volumes of users efficiently and said the company is working on a family of models to offer different tradeoffs of cost and performance: "wherever you want to be on that frontier... there'll be one of the models in the model family for you."
Efficiency, cost and serving billions
Cost and efficiency were presented as competitive advantages. Hassabis highlighted techniques such as distillation and other engineering work that let Google place advanced models into widely used products: "we've always tried to be on this frontier of cost to performance." Woodward added that integrating the model into high-usage surfaces lets Google collect signals and improve models over time.
Gemini Agent and personal productivity
One concrete near-term capability the guests discussed is an agent that can manage personal inboxes: read messages, propose replies and organize threads. Woodward said many users will notice immediate value from features that help organize and act on information, describing the inbox assistant as something he personally wanted to try: "one of the first things that I try when I get my hands on Gemini 3."
Style, persona and user relationships
On the model’s personality, Woodward described an intentional design toward a helpful, concise style. He framed Gemini as a "superpower" tool in a user’s toolbox and explained the team’s interest in measuring task completion: "how many tasks did we help you complete in your day?" Hassabis echoed that view, saying the goal is to make Gemini a workhorse for creativity, research and productivity rather than an emotional companion.
AGI timelines and expected breakthroughs
Asked whether Gemini 3 changed his AGI timeline, Hassabis reiterated the view he gave previously: progress is substantial but the path to artificial general intelligence still likely requires another one or two research breakthroughs. He framed Gemini 3 as "on track" with the team’s expectations and reaffirmed a five-to-ten-year horizon to AGI while noting remaining gaps in consistency, memory and world models.
it's on track of what I was expecting... still 5 to 10 years with one or two more perhaps breakthroughs required— Demis Hassabis
Safety and new capabilities
Both guests emphasized safety testing. Hassabis described Gemini 3 as the team’s "most thoroughly tested model so far," noting internal testing, work with safety institutes and external testers. They said improvements in tool use and function calling increase capability but also require heightened safety checks, particularly around risks such as cyber misuse.
Competition, market position and the AI bubble
When asked whether Google is back in the lead, Hassabis avoided declaring outright dominance and instead highlighted pace of progress and product integration: "the only important thing is your rate of progress... that's what we're focusing on." On the broader market, he suggested parts of the AI industry may feel bubbly — especially frothy early-stage funding — while arguing Google’s combination of near-term monetization and long-term platform opportunities positions the company to succeed in either scenario.
Everyday demos and immediate use cases
For an accessible demo, the guests suggested consumers try Gemini 3’s imagery and editing features on a phone (selfie editing and creative image generation) and then explore the app’s other capabilities such as interactive tutorials and agentic coding. They presented those demos as illustrative of the kinds of creative and productivity wins users might see quickly.
References and related links:
Google blog: A new era of intelligence with Gemini 3. (blog.google)
Hard Fork episode page: "Google's Gemini 3 Is Here: A Special Early Look" (Nov 18, 2025). (iheart.com)
TechCrunch coverage of the Gemini 3 launch. (techcrunch.com)
Episode listing and synopsis on Listen Notes. (listennotes.com)
Explore more exclusive insights at nextfin.ai.
