Developer Keynote in under 10 minutes

Year: 2026 · ▶ Watch on YouTube

Brad Calder (VP/GM, Google Cloud Platform and Technical Infrastructure) · Richard Seroter (Chief Evangelist) · Emma Twersky (Developer Advocate) · Mofi Rahman (Developer Advocate) · Ivan Nardini (Staff Software Engineer) · Casey West (Developer Advocate) · Jack Wotherspoon (Developer Advocate) · Lucia Subatin (Developer Advocate) · Megan O’Keefe (Developer Advocate) · Bobby Allen (Developer Advocate) · Ines Envid (Developer Advocate) · Jason Davenport (Developer Advocate) · Ankur Kotwal (Head of Product, Generative AI) · Yinon Costica (Co-Founder & VP of Product)

Switch language → zh

Segments (12)

  • 00:00:04 · Introduction — Brad Calder
    • Introduces the Gemini Enterprise Agent Platform and the keynote’s central demo: building an agentic marathon planner.
  • 00:00:29 · Introducing the Agentic Marathon Simulator — Richard Seroter, Emma Twersky
    • Explains the concept of the marathon simulator and its three core agents: Planner, Evaluator, and Simulator.
  • 00:01:13 · Demo 1: Build agents with Agent Platform — Mofi Rahman
    • Demonstrates building the ‘Planner agent’ by defining its tools (Google Maps MCP) and skills (Maps, GIS, Race Director).
  • 00:02:18 · Demo 2: Creating multi-agent systems — Ivan Nardini, Casey West
    • Shows how multiple agents collaborate with real-time evaluation and dynamic UI generation to prevent agent drift.
  • 00:02:55 · Enhancing agents with memory — Richard Seroter, Emma Twersky
    • Introduces Agent Platform Sessions and Memory Bank to give agents stateful intelligence and long-term memory.
  • 00:03:14 · Demo 3: Creating a data pipeline for RAG — Jack Wotherspoon, Lucia Subatin
    • Demonstrates using a Data Engineering agent to build a RAG pipeline with Lightning Engine for Spark, Document AI, and AlloyDB.
  • 00:04:22 · Demo 4: Debugging agents at scale — Megan O’Keefe
    • Uses Agent Observability and Gemini Cloud Assist to trace, diagnose, and receive a proactive code fix for an agent error.
  • 00:05:31 · Demo 5: Intent to infrastructure with Gemini Cloud Assist — Bobby Allen
    • Demonstrates using natural language to have Gemini Cloud Assist modify infrastructure-as-code, moving a service from Cloud Run to GKE.
  • 00:06:11 · Demo 6: Build and share no-code agents — Ines Envid, Jason Davenport
    • Shows how to create a complex, multi-sub-agent system using the no-code agent builder in Gemini Enterprise.
  • 00:07:09 · Demo 7: Securing agents — Ankur Kotwal
    • Demonstrates applying a security policy via Agent Gateway to restrict an agent’s access based on its Agent Identity.
  • 00:08:16 · Customer Story: Wiz — Yinon Costica
    • Wiz shows how their platform visualizes the agent architecture, finds security risks, and uses its own agent to suggest code fixes.
  • 00:09:01 · Conclusion — Richard Seroter, Emma Twersky
    • Recaps the Gemini Enterprise Agent Platform’s capabilities and provides a QR code to access the developer solution.

Products Announced (9)

  • 00:00:07 · Gemini Enterprise Agent Platform (Announced)
    • Unified platform to build, scale, govern, and optimize agents · Supports both high-code (ADK) and no-code (Gemini Enterprise) development · Integrates Observability, Security, and Memory
    • Available for developers to try.
  • 00:01:20 · Google Cloud MCP Server for Google Maps (Available)
    • Allows agents to access Google Maps tools securely · Connects agents to first-party Google tools · Managed service for tool access
  • 00:02:47 · Agent Registry (Announced)
    • Central repository for discovering and connecting agents · Enables agent-to-agent (A2A) communication · Manages agent metadata and endpoints
  • 00:02:56 · Agent Platform Sessions & Memory Bank (Announced)
    • Provides stateful intelligence for agents · Allows agents to learn from prior events · Manages context and long-term memory
  • 00:03:59 · Lightning Engine for Apache Spark (Available)
    • Optimized engine for running Spark jobs · Described as ‘lightning fast’ · Used for data processing in RAG pipelines
  • 00:04:31 · Agent Observability (Announced)
    • Provides granular instrumentation for agents · Allows visualization of execution paths and tool calls · Based on OpenTelemetry (OTel) compliant telemetry
  • 00:04:37 · Gemini Cloud Assist (Expanded)
    • Agentic cloud operator for design, run, secure, and optimize · Performs investigations and suggests proactive fixes · Modifies infrastructure-as-code from natural language
  • 00:07:29 · Agent Identity (Announced)
    • Assigns a unique cryptographic ID to every agent · Enables well-defined, traceable, and auditable authorization policies · Foundation for agent security
  • 00:07:39 · Agent Gateway (Announced)
    • Centralized location to manage agent policies · Mediates agent-to-agent and agent-to-tool communication · Enforces security policies defined with Agent Identity

Customer Stories (1)

  • 00:08:16 · Wiz — Cloud security platform that uses its own AI agent (‘Wiz Green Agent’) to analyze the security posture of the marathon planner application, identify an attack path, and suggest prioritized code fixes to remediate the vulnerabilities.

Demos (7)

  • 00:01:13 ✓ · Build agents with Agent Platform — Mofi Rahman
    • Building the ‘Planner agent’ by defining its tools (Google Maps MCP) and skills (Maps, GIS, Race Director) in code, resulting in a rendered marathon route.
  • 00:02:27 ✓ · Creating multi-agent systems — Casey West, Ivan Nardini
    • A multi-agent system where a Planner agent collaborates with an Evaluator sub-agent, featuring real-time evaluation and dynamic UI generation to plan a marathon.
  • 00:03:24 ✓ · Creating a data pipeline for RAG — Lucia Subatin, Jack Wotherspoon
    • Using a Data Engineering agent to create a declarative data pipeline, process documents with Spark and Document AI, and store embeddings in AlloyDB for RAG.
  • 00:04:46 ✓ · Debugging agents at scale — Megan O’Keefe
    • Using Agent Observability to trace a latency alert to a specific error, then using Gemini Cloud Assist to investigate the root cause and generate a proactive code fix.
  • 00:05:38 ✓ · Intent to infrastructure with Gemini Cloud Assist — Bobby Allen
    • Using a natural language prompt in an IDE to have Gemini Cloud Assist convert a Cloud Run YAML configuration to a GKE deployment configuration.
  • 00:06:24 ✓ · Build and share no-code agents — Ines Envid, Jason Davenport
    • Creating a ‘Supply Chain Agent’ with multiple sub-agents using the visual, no-code builder in Gemini Enterprise, and grounding it with a document from Google Drive.
  • 00:07:44 ✓ · Securing agents — Ankur Kotwal
    • An attempt to increase a budget via the agent is blocked by a new security policy enforced by the Agent Gateway, demonstrating role-based access control.

Notable Quotes (5)

  • 00:00:26 — Brad Calder:

    By building an agentic marathon planner.

  • 00:00:51 — Richard Seroter:

    Agent systems simulate actual behavior instead of hard-coding those behaviors or relying purely on probabilities.

  • 00:06:08 — Bobby Allen:

    This is our vision of an autonomous cloud.

  • 00:07:20 — Richard Seroter:

    We need to shift down instead. That means smarter platforms.

  • 00:09:28 — Richard Seroter:

    Our agent platform makes the process of building and running these agents the best in any cloud.

Visual Signals

On-screen (15)

  • 00:00:07 · Gemini Enterprise Agent Platform: Build, Scale, Govern, Optimize
    • Introduces the core product and its four pillars, structuring the entire keynote.
  • 00:00:45 · We built an agentic marathon simulator
    • Sets the context for the keynote’s central, recurring demonstration.
  • 00:00:58 · Diagram of Planner agent, Evaluator agent, and Simulator agent
    • Visually explains the multi-agent architecture of the demo application.
  • 00:01:15 · Build agents with Agent Platform
    • Title card for the first major demo segment.
  • 00:02:25 · Creating multi-agent systems
    • Title card for the second demo segment, focusing on agent collaboration.
  • 00:02:59 · Agent Platform Sessions & Memory Bank
    • Highlights the products responsible for making agents stateful.
  • 00:03:17 · Enhancing agents with memory
    • Title card for the third demo segment, focusing on RAG and data pipelines.
  • 00:04:31 · Agent Observability
    • Introduces the core product for debugging and monitoring agent behavior.
  • 00:04:37 · Gemini Cloud Assist
    • Introduces the agentic operator for managing cloud resources and code.
  • 00:06:17 · Build and share no-code agents
    • Title card for the demo on using Gemini Enterprise’s no-code builder.
  • 00:07:23 · Shift down
    • Introduces Google’s new paradigm for platform responsibility in security, contrasting it with ‘shift left’.
  • 00:07:44 · Securing agents
    • Title card for the security-focused demo segment.
  • 00:08:16 · WIZ logo
    • Introduces the customer showcase with Wiz, a cloud security partner.
  • 00:09:03 · Google Cloud Next '26
    • Closing branding for the event.
  • 00:09:07 · Explore the Developer Solution (with QR code)
    • Provides a direct call to action for the audience to access code, labs, and more information.

Stage (5)

  • 00:00:04 · Brad Calder walks on stage to begin the keynote.
  • 00:00:29 · Richard Seroter and Emma Twersky appear on stage.
  • 00:01:13 · The stage transitions to a single demo pod with Mofi Rahman.
  • 00:02:18 · The stage transitions to a demo pod with two presenters, Ivan Nardini and Casey West.
  • 00:09:45 · Richard Seroter and Emma Twersky conclude the keynote and the audience applauds.

Visual demos (7)

  • 00:01:19 · Code and diagrams for the Planner Agent
    • A diagram of the Planner Agent’s components (Instructions, Skills, Tools) is shown next to Python code in an IDE for connecting to the Maps MCP server and defining skills in Markdown files.
  • 00:02:27 · Multi-agent system UI
    • A chat-like UI shows a prompt to plan a marathon. The UI then displays a sequence of tool calls (‘plan_marathon_route’, ‘assess_traffic_impact’) while a 3D map of Las Vegas is dynamically rendered with the proposed route.
  • 00:03:50 · Data Engineering Agent in IDE
    • An IDE shows a prompt given to a ‘Data Engineering agent’ which then generates multiple files (pipeline.yaml, deployment.yaml) for a data pipeline. SQL code in AlloyDB shows the resulting embeddings.
  • 00:05:00 · Agent Observability and Cloud Assist
    • A Google Cloud Monitoring dashboard shows a ‘Request Latency Alert’. A detailed trace view is opened, showing an ‘ERROR’ on an ‘invoke_agent.simulator_agent’ span. A chat with Gemini Cloud Assist shows the agent identifying the root cause and generating a code diff to fix it.
  • 00:06:24 · Gemini Enterprise No-Code Agent Builder
    • A visual drag-and-drop style interface shows a main ‘Supply Chain Agent’ connected to four sub-agents (Waste Management, Supply Ordering, etc.). A Google Drive file is added as ‘Knowledge’ to the agent.
  • 00:07:55 · Agent Security Policy in Action
    • A user asks the agent to increase the budget. The agent’s response is ‘Access Restricted’ with a message explaining it is not authorized to change budget allocations, demonstrating the security policy.
  • 00:08:25 · Wiz Security Graph
    • The Wiz UI displays a ‘Security Graph’ showing the full architecture of the marathon app, highlighting an attack path from the ‘Internet’ to the ‘Planner’ agent and its underlying service account.

Key Topics

Agentic AI · Multi-agent Systems · Gemini Enterprise Agent Platform · Developer Experience · AI-driven Development · No-Code AI Development · AI Observability · AI Security · Autonomous Cloud · Retrieval-Augmented Generation (RAG) · Infrastructure as Code (IaC) · Stateful AI · AI Governance

Takeaways

  • Google is launching the Gemini Enterprise Agent Platform, a comprehensive, end-to-end solution for building, scaling, governing, and optimizing production-ready AI agents.
  • The platform uniquely supports both high-code (ADK, Agent Runtime) and no-code (Gemini Enterprise) agent development, enabling them to collaborate within the same system via an Agent Registry.
  • A major focus is on operational readiness, with built-in tools for stateful memory (Memory Bank), multi-agent evaluation, and debugging (Agent Observability).
  • Security and governance are addressed through a ‘shift down’ approach, where the platform takes more responsibility using Agent Identity and Agent Gateway to enforce fine-grained, auditable policies.
  • Gemini Cloud Assist is evolving into an ‘agentic cloud operator’ that can understand intent from natural language to modify infrastructure, diagnose issues, and proactively fix code, pushing towards the vision of an ‘autonomous cloud’.
  • The platform is deeply integrated with the Google Cloud ecosystem, leveraging services like Google Maps, Document AI, AlloyDB, and Spark for building powerful, data-grounded agents.
  • The entire keynote was structured around a single, cohesive demo of building a ‘marathon planner,’ showcasing each component of the Agent Platform in a practical, step-by-step manner.