Cloud Next ‘26 Developer Keynote

Year: 2026 · ▶ Watch on YouTube

Brad Calder (President, GCP and SRE, Google Cloud) · Richard Seroter (Chief Evangelist, Google Cloud) · Emma Twersky (Developer Relations Engineer, Google Cloud) · Mofi Rahman (Developer Relations Engineer, Google Cloud) · Ivan Nardini (Developer Relations Engineer, Google Cloud) · Casey West (Architecture Advocate, Google Cloud) · Lucia Subatin (Developer Relations Engineer, Google Cloud) · Jack Wotherspoon (Developer Relations Engineer, Google Cloud) · Megan O’Keefe (Senior Staff Developer Advocate, Google Cloud) · Bobby Allen (Developer Relations Engineer, Google Cloud) · Ines Envid (Developer Relations Engineer, Google Cloud) · Jason Davenport (Developer Relations Engineer, Google Cloud) · Ankur Kotwal (Developer Relations Engineer, Google Cloud) · Yinon Costica (Co-Founder & VP of Product, Wiz)

Switch language → zh

Segments (10)

  • 00:02:28 · Introduction to the Gemini Enterprise Agent Platform — Brad Calder
    • Introduces the Gemini Enterprise Agent Platform for building, scaling, governing, and optimizing autonomous AI agents, setting the stage for the keynote’s central demo.
  • 00:06:02 · Keynote Introduction and Developer Stories — Richard Seroter & Emma Twersky
    • The hosts kick off the main keynote, share inspiring developer projects built with Google AI, and introduce the marathon simulator demo.
  • 00:13:27 · Demo 1: Build agents with Agent Platform — Mofi Rahman
    • Demonstrates building the foundational ‘Planner Agent’ for the marathon simulator using the Agent Development Kit (ADK), skills, and tools.
  • 00:20:41 · Demo 2: Creating multi-agent systems — Ivan Nardini & Casey West
    • Expands the system by creating an ‘Evaluator’ sub-agent and a ‘Simulator’ agent, showing how they communicate using the A2A protocol and Agent Registry.
  • 00:28:43 · Demo 3: Enhancing agents with memory — Lucia Subatin & Jack Wotherspoon
    • Demonstrates adding stateful intelligence to agents using Agent Platform Sessions, Memory Bank, and Retrieval-Augmented Generation (RAG) with AlloyDB.
  • 00:35:42 · Demo 4: Debugging agents at scale — Megan O’Keefe
    • Troubleshoots a simulated production error using Agent Observability and Gemini Cloud Assist to identify the root cause and suggest a code fix.
  • 00:42:07 · Demo 5: Intent to infrastructure with Gemini Cloud Assist — Bobby Allen
    • Uses natural language with Gemini Cloud Assist to upgrade the runner agent’s infrastructure from Cloud Run to a GKE deployment with a custom Gemma model.
  • 00:48:51 · Demo 6: Build and share no-code agents — Ines Envid & Jason Davenport
    • Shows how to create a no-code ‘Supply Chain Agent’ in Gemini Enterprise and share it via the Agent Registry to collaborate with the existing marathon planner.
  • 00:55:56 · Demo 7: Securing agents — Ankur Kotwal & Yinon Costica
    • Demonstrates securing the multi-agent system using Agent Gateway, Agent Identity, IAM policies, and the Wiz security platform to find and remediate vulnerabilities.
  • 01:06:00 · Conclusion and Call to Action — Richard Seroter & Emma Twersky
    • Summarizes the capabilities of the Gemini Enterprise Agent Platform and announces that the entire marathon simulator demo is open-sourced.

Products Announced (9)

  • 00:02:40 · Gemini Enterprise Agent Platform (New Platform)
    • A comprehensive platform to build, scale, govern, and optimize autonomous agents. · Includes components like ADK, Agent Runtime, Agent Registry, and Agent Gateway. · Enables multi-agent systems and shared context across applications.
    • Not mentioned
  • 00:03:04 · Gemini 3.1 (New Version)
    • Powers the Agent Platform. · Includes Pro and Flash models.
    • Available
  • 00:20:09 · A2A Protocol (Agent-to-Agent) (New Standard)
    • An open standard for seamless communication and collaboration between AI agents. · Donated to the Linux Foundation.
    • Open Standard
  • 00:20:17 · Agent Registry (New Feature)
    • A central directory where agents can be discovered and their capabilities advertised. · The ‘DNS of the internet of agents’.
    • Available
  • 00:23:12 · A2UI (Agent-to-User Interface) (New Standard)
    • An open standard for agent-driven interfaces. · Allows agents to dynamically generate UI components. · Moves beyond simple text-based responses.
    • Open Standard
  • 00:27:59 · Agent Platform Sessions & Memory Bank (New Features)
    • Provides stateful intelligence to agents over time. · Memory Bank allows agents to pull learnings from prior events. · Enables long-term context and personalization.
    • Available
  • 00:35:06 · Agent Observability (New Feature)
    • Delivers granular instrumentation for agent behavior. · Allows visualization of the full execution path using OpenTelemetry. · Integrates with Cloud Monitoring and Cloud Trace.
    • Available
  • 00:35:13 · Gemini Cloud Assist (New Product)
    • A personalized, agentic cloud operator. · Helps design, run, secure, and optimize agents and applications. · Can be used for troubleshooting and infrastructure modification via natural language.
    • Available
  • 00:56:12 · Wiz Integration (New Partnership)
    • Provides a security layer for agentic systems. · Scans code and infrastructure to build a security graph. · Uses ‘Red Agent’ for penetration testing and ‘Green Agent’ for remediation.
    • Available

Customer Stories (4)

  • 00:07:05 · DeepMind — Andy Coenen used Gemini to transform geographic data into nostalgic 90s-style isometric maps of NYC.
  • 00:07:27 · Cloudphysician — A Google Startup Accelerator participant in India built a multimodal video AI platform on Google Cloud to connect bedside teams with a central care team, reducing ICU mortality.
  • 00:08:01 · Cagatay Ulusoy (Google Developer Expert) — Built a generative UI-powered language learning app using Flutter, Firebase, and Gemini to study for his Finnish citizenship exam.
  • 00:08:35 · Waymo — Uses Genie 3 world models to recreate edge-case traffic situations to train their foundation models for self-driving.

Commitments (1)

  • 01:06:02 (Immediately) — Open-source every line of code shown in the keynote demo.

Demos (8)

  • 00:09:16 ✓ · Agentic Marathon Simulator — Multiple
    • A multi-part, continuous demo building a complex, multi-agent system to plan, simulate, debug, and secure a marathon event in Las Vegas.
  • 00:13:27 ✓ · Building the Planner Agent — Mofi Rahman
    • Using Agent Designer and the ADK to create a basic planner agent with instructions, skills (Maps, GIS), and tools (Google Maps MCP server).
  • 00:20:41 ✓ · Creating a Multi-Agent System — Ivan Nardini & Casey West
    • Creating an Evaluator sub-agent and a separate Simulator agent, connecting them via the A2A protocol and Agent Registry, and showing the dynamic A2UI.
  • 00:28:43 ✓ · Enhancing Agents with Memory and RAG — Lucia Subatin & Jack Wotherspoon
    • Adding stateful memory using Agent Platform Sessions and Memory Bank, and incorporating unstructured data (parking rules) via a RAG pipeline with AlloyDB.
  • 00:34:24 ✓ · Debugging a Production Failure — Megan O’Keefe
    • A simulated production failure (sirens, red screen) was diagnosed using Agent Observability, Cloud Monitoring, and Gemini Cloud Assist to find a context window limit issue and generate a code fix.
  • 00:42:07 ✓ · Intent-to-Infrastructure with Gemini Cloud Assist — Bobby Allen
    • Using a natural language prompt in an IDE to instruct Gemini Cloud Assist to migrate the ‘runner’ service from Cloud Run to GKE and deploy a custom Gemma 4 model.
  • 00:48:51 ✓ · Building and Sharing No-Code Agents — Ines Envid & Jason Davenport
    • Creating a ‘Supply Chain Agent’ in Gemini Enterprise using a no-code interface, grounding it with a Google Doc, and then using it in conjunction with the Marathon Planner agent.
  • 00:55:56 ✓ · Securing the Agentic System with Wiz — Ankur Kotwal & Yinon Costica
    • Using Agent Gateway policies to restrict agent access, and then using the Wiz security platform to discover and remediate a critical attack path involving prompt injection.

Notable Quotes (4)

  • 00:02:43 — Brad Calder:

    It allows you to build autonomous agents that proactively help users and complete tasks independently.

  • 00:47:16 — Bobby Allen:

    Don’t chat with me about change, partner with me to safely implement change. This is our vision of an autonomous cloud.

  • 00:55:08 — Richard Seroter:

    Shift left is often code for ‘developers should just do more’. It’s not sustainable… We need to shift down instead. That means smarter platforms.

  • 01:07:03 — Emma Twersky:

    Now let’s go build.

Visual Signals

On-screen (12)

  • 00:02:29 · Google Cloud Next '26
    • Establishes the event name and year.
  • 00:02:44 · Gemini Enterprise Agent Platform - Build, Scale, Govern, Optimize
    • The central product announcement of the keynote.
  • 00:03:24 · Agent Development Kit
    • Introduces the first pillar of the Agent Platform.
  • 00:13:29 · Build agents with Agent Platform
    • Title card for the first major demo segment.
  • 00:20:09 · A2A Protocol
    • Highlights the new open standard for agent-to-agent communication.
  • 00:23:12 · A2UI - A protocol for agent-driven interfaces
    • Highlights the new open standard for agents to generate user interfaces.
  • 00:27:59 · Agent Platform Sessions & Memory Bank
    • Shows the components for making agents stateful.
  • 00:35:06 · Agent Observability
    • Introduces the suite of tools for debugging agentic systems.
  • 00:42:24 · Intent to infrastructure with Gemini Cloud Assist
    • Title card for the infrastructure-as-code demo segment.
  • 00:45:39 · Hypothesis: Suboptimal Storage Configuration (GCS Fuse vs Lustre)
    • Shows the output of a Gemini Cloud Assist investigation, identifying a performance bottleneck.
  • 00:48:45 · QR code for goo.gle/buildagents Codelab
    • Provides a direct link for developers to try the demo themselves.
  • 01:06:09 · Explore the Developer Solution - QR code for the full open-source project
    • The main call to action, linking to the complete source code for the keynote demo.

Stage (6)

  • 00:00:00 · An energetic, abstract 3D animation intro video plays.
  • 00:02:19 · Brad Calder walks onto the main stage to applause.
  • 00:10:24 · The ‘magic stage’ rises from the floor, revealing two demo stations for the hosts.
  • 00:34:28 · The stage lighting turns red and a loud siren blares to simulate a production failure during the demo.
  • 00:49:04 · One of the hosts throws a white rally towel into the audience.
  • 01:07:05 · The hosts conclude the keynote, and confetti cannons fire as they walk off stage.

Visual demos (7)

  • 00:09:54 · An agentic marathon simulator is introduced, showing a 3D map of Las Vegas with a proposed race route.
    • A 3D rendered map of the Las Vegas strip with neon-lit buildings and a dotted line representing the marathon path.
  • 00:14:19 · Using Agent Designer to create the ‘Marathon Planner’ agent with natural language instructions.
    • A web-based UI for Agent Designer showing a flow-chart like canvas with an agent node and its configuration panel.
  • 00:23:28 · The A2UI protocol is used by the Evaluator agent to dynamically generate a UI card displaying the marathon plan’s scores.
    • A UI card with a title, score, and various metrics like ‘Safety Compliance’ and ‘Community Impact’ appears in the chat interface.
  • 00:37:14 · The Cloud Trace view is used to debug the agent’s execution flow.
    • A detailed waterfall chart showing nested spans for each tool call and LLM invocation, with one span highlighted in red indicating an error.
  • 00:43:28 · Gemini Cloud Assist in an IDE (Antigravity) is used to convert a Cloud Run YAML file to a GKE deployment YAML.
    • A chat interface within the IDE where a user provides a voice prompt, and the AI generates and modifies YAML configuration files.
  • 00:51:38 · The Gemini Enterprise Agent Designer is used to create a no-code ‘Supply Chain Agent’ by providing a prompt and grounding it with a Google Doc.
    • A visual flow diagram showing the main agent and its sub-agents for different tasks like Waste Management and Nutrition.
  • 01:00:40 · The Wiz Security Graph is used to visualize and remediate a critical attack path.
    • A node-and-edge graph showing connections from the internet to the planner agent, highlighting a vulnerability and its path to sensitive data.

Key Topics

Agentic AI · Multi-Agent Systems · Gemini Enterprise Agent Platform · Developer Experience · Infrastructure as Code (IaC) · AI-assisted Development · Observability and Debugging · Cloud Security · No-Code/Low-Code Development · Retrieval-Augmented Generation (RAG) · Stateful AI Agents · Open Standards (A2A, A2UI) · Google Kubernetes Engine (GKE) · Serverless Computing

Takeaways

  • Google is launching the Gemini Enterprise Agent Platform, a comprehensive suite of tools for building, scaling, governing, and optimizing production-ready autonomous AI agents.
  • The future of AI applications is multi-agent systems, where specialized agents collaborate to solve complex problems, communicating through open standards like A2A (Agent-to-Agent).
  • Google is introducing new open standards like A2A and A2UI (Agent-to-User Interface) to create a more interoperable and dynamic ecosystem for agentic AI.
  • Observability is critical for complex agentic systems. Agent Observability and Gemini Cloud Assist provide deep tracing and AI-powered debugging to quickly resolve production issues.
  • Security is being ‘shifted down’ to the platform. Agent Identity, Agent Gateway, and partnerships with companies like Wiz provide robust governance and security for agentic applications from code to cloud.
  • Developers can interact with infrastructure and build agents using a spectrum of tools, from high-code (ADK in an IDE) to no-code (Gemini Enterprise visual designer), catering to all skill levels.
  • Stateful intelligence is key for personalization and efficiency. The Agent Platform’s Sessions and Memory Bank allow agents to retain context and learn from past interactions.
  • The entire marathon simulator demo, showcasing all these technologies, is being open-sourced, providing a complete developer solution for the community to explore and build upon.