Cloud Next ‘24 Developer Keynote

Year: 2024 · ▶ Watch on YouTube

Richard Seroter (Chief Evangelist) · Chloe Condon (Chief Developer Relations Engineer) · Brad Calder (GM & VP) · Jason Davenport (Developer Advocate) · Femi Akinde (Product Manager) · Guillermo Rauch (CEO) · Kaslin Fields (Developer Advocate) · Josh Long (Spring Developer Advocate) · Chen Goldberg (VP & GM, Cloud Runtimes) · Steve McGhee (Reliability Advocate) · Charity Majors (Co-Founder & CTO) · Philipp Schmid (Engineering Manager)

Switch language → zh

Segments (11)

  • 00:00:08 · Introduction and Welcome — Richard Seroter & Chloe Condon
    • The hosts welcome the audience to the Developer Keynote and set the stage for how Gemini is transforming the developer experience.
  • 00:01:28 · Gemini 1.5 Pro Demo: Analyzing Last Year’s Keynote — Richard Seroter & Chloe Condon
    • A demonstration of Gemini 1.5 Pro’s multimodal capabilities by analyzing the video of the 2023 Developer Keynote to extract insights.
  • 00:03:32 · Better Coding with Gemini Code Assist — Brad Calder
    • Introduction to Gemini for Google Cloud, focusing on Gemini Code Assist and its 1 million token context window for large-scale code tasks.
  • 00:05:06 · Better Ops with Gemini Cloud Assist — Brad Calder & Jason Davenport
    • A live demo of Gemini Cloud Assist being used to troubleshoot a production issue in the Google Cloud Console by analyzing alerts and logs.
  • 01:12:08 · Better Data with Gemini in BigQuery — Brad Calder & Jason Davenport
    • Demonstration of building a real-time analytics application using Gemini in BigQuery with continuous queries to analyze social media trends.
  • 01:18:18 · Build with Gemini: Creating a Generative UI App — Chloe Condon, Femi Akinde, & Guillermo Rauch
    • A walkthrough of building an AI-powered travel recommender app, moving from idea in Google AI Studio to a full Next.js application with Generative UI.
  • 01:28:27 · Run with Gemini: Production-Grade AI Apps — Richard Seroter, Kaslin Fields, & Josh Long
    • Discussion and demos on running GenAI applications at scale using GKE, AlloyDB for RAG, and building fast, scalable APIs with Spring Boot and Spring AI.
  • 01:41:42 · Better Platforms with Gemini — Richard Seroter & Chen Goldberg
    • Chen Goldberg showcases how Gemini Cloud Assist simplifies creating and managing complex application architectures on Cloud Run and GKE.
  • 01:49:26 · Operate with Gemini: AI in Observability — Chloe Condon, Steve McGhee, & Charity Majors
    • Experts discuss and demonstrate how to operate and debug GenAI applications using Vertex AI for prompt management and Honeycomb for observability.
  • 02:01:11 · Partner Showcase: Hugging Face Integration — Richard Seroter & Philipp Schmid
    • A demonstration of the deep integration between Hugging Face and Google Cloud, allowing for easy training and one-click deployment of models to Vertex AI and GKE.
  • 02:03:50 · Closing and Key Takeaways — Richard Seroter, Chloe Condon, & Jason Davenport
    • The hosts summarize the keynote’s main themes by using Gemini 1.5 Pro to analyze the keynote itself in real-time.

Products Announced (6)

  • 00:02:29 · Gemini 1.5 Pro (Generally Available)
    • 1 million token context window · Multimodal capabilities (video, audio, text, code) · Long-context reasoning
    • Available in Vertex AI
  • 00:04:14 · Gemini Code Assist (Updated)
    • 1 million token context window integration · Comprehensive code reviews · Large-scale codebase understanding for adding features and upgrades
    • Available
  • 00:05:08 · Gemini Cloud Assist (New, Private Preview)
    • AI assistance across the entire app lifecycle · Context-aware troubleshooting and optimization · Natural language interface for managing cloud resources
    • Private Preview
  • 01:12:08 · Gemini in BigQuery (New Features)
    • Continuous queries for real-time analytics · Multimodal data analysis · Natural language to SQL generation
    • Available
  • 01:45:45 · Ray on GKE (New Support)
    • Simplified deployment via a single checkbox · Managed service for the Ray framework · Enables distributed AI/ML training and serving on GKE
    • Available
  • 01:51:14 · Vertex AI Prompt Management & Evaluation (New Features)
    • Side-by-side prompt comparison · Ground truth evaluation with ROUGE and BLEU scores · Version control and history for prompts
    • Available

Benchmarks Shown (1)

  • 02:29:58 · AlloyDB Performance: 4x faster
    • Standard PostgreSQL

Demos (9)

  • 00:01:28 ✓ · Gemini 1.5 Pro Video Analysis — Chloe Condon
    • Using Gemini 1.5 Pro in Google AI Studio to analyze the 75-minute video of the previous year’s keynote and answer specific questions about its content.
  • 00:06:05 ✓ · Gemini Cloud Assist Troubleshooting — Jason Davenport
    • Using the Gemini side panel in the Google Cloud Console to identify the root cause of an application error by asking natural language questions about alerts, logs, and firewall rules.
  • 01:12:08 ✓ · Gemini in BigQuery Real-Time Analytics — Jason Davenport
    • Building a real-time data pipeline to analyze social media posts. It used Gemini to generate SQL for sentiment and topic extraction, and then enabled continuous queries to stream results to Pub/Sub.
  • 01:19:19 ✓ · Generative UI with Next.js and Vercel AI SDK — Guillermo Rauch
    • Building an interactive travel booking chatbot that uses Gemini and the Vercel AI SDK to render React components (like flight lists and seat pickers) directly in the chat interface instead of just text.
  • 02:32:45 ✓ · GenAI API with Spring Boot — Josh Long
    • Rapidly creating a production-ready, AI-enabled Java API using Spring Initializr, Spring AI, and GraalVM to build a native image that starts instantly and performs a RAG-based product recommendation.
  • 02:42:38 ✓ · Cloud Run Application Canvas — Chen Goldberg
    • Using a new visual canvas in Cloud Run to describe a desired application architecture (e.g., ‘web app with database’) in natural language, which Gemini Cloud Assist then diagrams and provisions.
  • 02:55:19 ✓ · Observability with Honeycomb — Charity Majors
    • Using Honeycomb’s Query Assistant (powered by Gemini) to debug a GenAI application by asking natural language questions to identify latency issues and root causes of errors in the RAG pipeline.
  • 03:01:46 ✓ · Hugging Face to Google Cloud Deployment — Philipp Schmid
    • Demonstrating the one-click ‘Deploy on Google Cloud’ feature from the Hugging Face Model Hub to deploy the Gemma model directly to a Vertex AI endpoint.
  • 03:04:28 ✓ · Keynote Self-Analysis — Jason Davenport
    • As a closing act, the keynote video itself was fed into Gemini 1.5 Pro to generate a summary of the key products and takeaways from the presentation.

Notable Quotes (3)

  • 02:53:41 — Steve McGhee:

    May all your incidents be novel.

  • 03:05:27 — Richard Seroter:

    From coding, to running platforms, to operations, software is simply better with Gemini.

  • 02:24:18 — Guillermo Rauch:

    We believe this is the future, and we call this Generative UI.

Visual Signals

On-screen (14)

  • 00:00:04 · Title Card: Developer keynote
    • Sets the theme and title of the event.
  • 00:00:40 · Logo: Google Cloud Next
    • Brands the event.
  • 00:01:15 · Postcard graphic: Greetings from LEGACY LAND '23
    • A visual callback to the theme of the previous year’s keynote, setting up a comparison.
  • 00:02:29 · Slide: Gemini 1.5 Pro, 1M tokens
    • Highlights the key capability of the model being demonstrated.
  • 00:03:09 · Slide: Gemini for Google Cloud - Better coding, Better platforms, Better ops
    • Outlines the three core pillars of the keynote’s narrative.
  • 00:04:15 · Slide: Gemini Code Assist - AI assistance for developing apps
    • Introduces the first major product focus.
  • 00:05:09 · Slide: Gemini Cloud Assist - AI assistance across app lifecycle
    • Introduces the second major product focus.
  • 01:14:34 · Slide: Gemini in BigQuery
    • Introduces the data analytics segment.
  • 01:18:24 · Slide: Build, Run, Operate
    • A transition slide reinforcing the keynote’s structure.
  • 01:18:30 · Slide: Build with Gemini
    • Title card for the ‘Build’ section of the keynote.
  • 02:28:25 · Slide: Run with Gemini
    • Title card for the ‘Run’ section of the keynote.
  • 02:49:14 · Slide: Operate with Gemini
    • Title card for the ‘Operate’ section of the keynote.
  • 02:53:45 · Quote Slide: 'May all your incidents be novel' - SRE proverb
    • A well-known quote in the operations community, setting the stage for the next topic.
  • 03:01:39 · Partnership Logo: Google Cloud | Hugging Face
    • Visually represents the partnership being discussed.

Stage (9)

  • 00:00:14 · Hosts Richard Seroter and Chloe Condon walk out from the audience onto the stage.
  • 00:03:42 · Brad Calder, GM & VP of Google Cloud, walks on stage.
  • 00:05:58 · Jason Davenport, Developer Advocate, appears at a demo pod on stage.
  • 01:18:42 · Femi Akinde, Product Manager, walks on stage to join Chloe Condon at the demo pod.
  • 02:02:01 · Guillermo Rauch, CEO of Vercel, walks on stage.
  • 02:48:49 · Kaslin Fields, Developer Advocate, walks on stage.
  • 02:52:22 · Josh Long, Spring Developer Advocate, walks on stage.
  • 02:53:50 · Charity Majors, Co-Founder & CTO of honeycomb.io, walks on stage.
  • 03:01:29 · Philipp Schmid, Engineering Manager at Hugging Face, walks on stage.

Visual demos (8)

  • 00:02:09 · Google AI Studio UI
    • A prompt with multiple tasks for video analysis is shown, along with the resulting markdown-formatted analysis of the 2023 keynote video.
  • 00:06:30 · Google Cloud Console with Gemini side panel
    • The user interacts with the Gemini chat panel, asking it to show recent alerts for ‘cymbal-shop-app’. It then navigates through incident details and log explorer, using Gemini to explain log entries and suggest troubleshooting commands.
  • 01:13:12 · BigQuery Studio UI
    • A user queries social media posts, then uses the ‘Help me code’ feature to generate a complex SQL statement with a prompt to extract sentiment and topics using an LLM. The query is then modified to become a continuous query that exports data to Pub/Sub.
  • 02:06:04 · Next.js Chatbot Application
    • A web-based chatbot UI is shown. When a user asks to book a flight, the UI dynamically renders a list of flights (a React component). When a flight is selected, it renders a seat-picker component, demonstrating Generative UI.
  • 02:36:34 · Spring Initializr and IntelliJ IDEA
    • The Spring Initializr website is used to bootstrap a new Spring Boot project. The code is then shown in an IntelliJ IDE, where a native image is compiled using GraalVM, and the resulting fast-starting API is called from the terminal.
  • 02:43:02 · Cloud Run Application Canvas
    • A new visual interface in Cloud Run where a user types ‘Web app with database’ into a prompt. A diagram with a ‘Service’ and ‘CloudSQL’ block appears, which can then be configured and deployed.
  • 02:55:19 · Honeycomb.io UI
    • A dashboard showing SLOs and a ‘BubbleUp’ view that automatically highlights dimensions (like ‘error’) that are most different between failing and successful requests. The user then uses the natural language Query Assistant to investigate further.
  • 03:01:59 · Hugging Face and Vertex AI UI
    • The Gemma model page on Hugging Face is shown, with a ‘Deploy’ dropdown that has a ‘Google Cloud’ option. Clicking it leads to a Vertex AI ‘Deploy from Hugging Face’ page, pre-filled with the model details for one-click deployment.

Key Topics

Generative AI · Developer Experience · LLM Operations (LLMOps) · Cloud Native Development · RAG (Retrieval-Augmented Generation) · Multimodal AI · Prompt Engineering · Serverless Computing · Kubernetes (GKE) · Observability · Data Analytics · Application Frameworks · Spring Framework · Next.js · AI-assisted Development

Takeaways

  • Gemini is being deeply integrated across the entire Google Cloud ecosystem to assist developers in coding, running platforms, and operating applications.
  • The 1 million token context window in Gemini 1.5 Pro unlocks new capabilities for understanding and reasoning over massive amounts of information, like entire codebases or long videos.
  • Gemini Cloud Assist acts as a context-aware AI partner within the Google Cloud Console, simplifying complex tasks like architecture design and production troubleshooting.
  • Google is making it easier to build RAG and real-time AI applications through native vector database capabilities in products like AlloyDB and continuous queries in BigQuery.
  • The concept of ‘Generative UI’, showcased with Vercel’s Next.js AI SDK, represents a shift from text-only chatbots to rich, component-based interactive AI experiences.
  • Partnerships with open-source leaders like Hugging Face, Vercel, and Broadcom (Spring) are central to Google’s strategy, making it simpler to use popular frameworks and models on Google Cloud.
  • Operating GenAI applications requires new approaches to observability and prompt management, which are being addressed by tools in Vertex AI and from partners like Honeycomb.
  • Google is focused on improving the performance and efficiency of AI workloads through optimizations in GKE and Cloud Run, including support for TPUs, GraalVM native images, and frameworks like Ray.