Building in the Gemini Era with Google DeepMind

1:45pm - 2:04pm | Building in the Gemini Era with Google DeepMind

Speakers: Kat Kampf & Ammaar Reshi (both Google)

Speaker Profiles: Kat Kampf | Ammaar Reshi

Bio: Kat Kampf - Product Manager, Google; Ammaar Reshi - Product & Design Lead, Google

Topic: (tba on talk content)

Notes

• exciting week for deepmind
• innovating for many years
• gemini 3 pro
• build anything
  • ui and sensibility
  • agentic tool calling
• nanonanana
  • has search
  • can ask it and it will give e.g. cooking instructions
  • e.g. change the focus
  • and big stuff
• ai.studio/build
  • build experience
  • free to use
  • bunch of ai trips
  • webcam on tennis spring
  • shared application — build and sharing
  • use google search grounding
• giving you ideas during the load screen about new ideas
  • like the proactive stuff that jules was talking about yesterday
• vibe coding a comix book
  • vibe coding and winging the presentation
  • very fun way of building the story, gemini 3
  • includes a choose your own adventure
• design sensibilities
  • no more purple gradients
  • shader animations
  • no cyberpunk shit
  • “no more groking their way around figma”
• sticker model
  • add an api key
  • add different words for laptop stickers
• using aistudio to build aistudio
  • it has an export to antigravity
• Making video games
  • have a bot now
  • make a start screen
  • all these are front end react app
• full backend support and runtime support
  • we dont want you to think about those details
• live demo of a one shot networking game
  • too many people in the room but it looked like it was working
• first gen where anyone can build software

Slides

Slide: 13-48

Key Point: Gemini 1.5 Pro achieves the highest benchmark performance (74.2%) across diverse tasks while being competitively priced, demonstrating superior capabilities in reasoning, coding, mathematics, and multimodal tasks compared to other leading models.

Literal Content:

• Comprehensive benchmark comparison table showing Gemini 1.5 Pro vs Sonnet 3.5, Claude 3.5, and GPT-4o mini
• 20 different benchmarks listed including: Humanity’s Last Exam, ARC-AGI, GPQA Diamond, JME 2025, MATHSS00, COMMMLU-Pro, and more
• Bar chart showing cost comparison: Gemini 1.5 Pro (74.2%, $320), Sonnet 3.5 (70.6%, $279), Opus 4 (66.4%, $160), GPT-4 (65.0%, $140), GPT-4 mini (59.8%, $16)
• Footer references methodology at deepmind.google/models/evals-methodology/gemini-1.5-pro

Slide: 13-49

Key Point: Demonstrates Gemini’s ability to generate culturally specific, detailed instructional content with visual elements - showcasing multimodal generation capabilities that combine text, images, and structured information presentation.

Literal Content:

• Title: “World knowledge - HOW TO MAKE ELAICHI CHAI (CARDAMOM TEA)”
• Decorative floral border design with pink/red flowers
• Ingredients shown with icons: Water, Milk, Tea Luxury Powder, Green Cardamom, Sugar (Optional), Ginger
• 5-step process with illustrated diagrams:
  1. Prepare Cardamom & Water (Crush 2-3 cardamom pods, bring 1 cup water to boil)
  2. Add Tea & Spices (Add 1-2 tsp tea leaves, crushed cardamom, optional ginger, simmer 2-3 minutes)
  3. Pour Milk & Sugar (Pour 1 cup milk, add sugar to taste, bring to boil again)
  4. Simmer & Steep (Lower heat, simmer another 2-3 minutes for flavor development)
  5. Strain & Serve (Strain hot chai into cups, enjoy!)
• Tips section

Slide: 13-53

Key Point: Gemini provides a comprehensive development platform with AI-powered features including app generation from natural language, image generation, intelligence integration, and app idea discovery - positioning it as an end-to-end solution for AI-powered application development.

Literal Content:

• Title: “Build your ideas with Gemini”
• Top section shows prompt interface with model selector showing “Gemini 3 Pro Preview” with “Build →” button
• Middle section: “Supercharge your apps with AI”
• Four feature cards in carousel:
  • Google Maps data integration
  • Generate images from Gemini
  • Connect intelligence in your app
  • AI personal status features
• Bottom section: “Discover and remix app ideas” with app preview cards

Slide: 13-58

Key Point: Showcases Gemini’s ability to generate personalized, themed creative content based on user interests and personas - demonstrating practical application of AI for customization and design tasks that appeal to individual preferences.

Literal Content:

• Title: “Design Your Vibe.”
• Subtitle: “Create a unique laptop sticker pack. Generate infinite sticker aesthetic, instantly! Describe your interests or search for a persona.”
• Two input fields:
  • “What are you into? (e.g., coding, retro gaming)”
  • “Who is this for? (e.g., Emma, a web developer)”
• Two example outputs with photos of laptops:
  • MacBook with corgi/dog stickers and coding-themed stickers - labeled “Generate an animal pack”
  • MacBook covered with developer-themed stickers - labeled “Generate Cover”

Slide: 14-02

Key Point: Demonstrates Gemini’s ability to generate and modify game code interactively, showing live code editing for game mechanics (acceleration/deceleration) and the resulting visual output - illustrating practical code generation for complex interactive applications despite encountering some errors.

Literal Content:

• Browser window showing Google AI Studio interface
• Left panel shows TypeScript code with game mechanics variables:
  • const ACCELERATION = 0.000;
  • const DECELERATION = 0.150;
• “Checkpoint” and “View diff”/“Restore” buttons visible
• Error message showing internal error
• “Retry” button
• Right panel shows “Vice City Racer” game with retro synthwave aesthetic
• Game shows “VICE CITY SYNTH RACING” title, position “1 / 11”, “Online Players: 13”
• Multiplayer status and 3D racing track with geometric shapes, neon colors

Slide: 14-03

Key Point: A bold, aspirational statement positioning the current era as a fundamental shift in software development - where AI tools are democratizing programming and enabling non-engineers to build software, representing a paradigm shift in who can create technology.

Literal Content:

• Pink background with white text
• Single sentence centered on slide: “We’re the first generation of engineers building tools for a world where anyone can build software”