Google for Developers’ cover photo
Google for Developers

Google for Developers

Technology, Information and Internet

Mountain View, CA 4,058,235 followers

Join a community of creative developers and learn how to use the latest in technology—from AI and cloud, to mobile & web

About us

Discover the latest technologies, resources, events, and announcements to help you build smarter and ship faster. Explore more at developers.google.com

Website
http://developers.google.com
Industry
Technology, Information and Internet
Company size
10,001+ employees
Headquarters
Mountain View, CA
Specialties
coding, engineering, firebase, android, cloud, web development, and mobile development

Updates

  • In the #GeminiLiveAgentChallenge, we challenged developers to build the next generation of agents that can help you see, hear, speak, and create with Gemini and Google Cloud. The results for projects that truly moved beyond the text box are in. 🏆 And the grand prize winner is…ORION! ORION, or Operating Room Intelligent Orchestration Node, is a voice-directed surgical co-pilot for robotic surgery. Surgeons can speak naturally and instantly receive answers, live data on display, and real-time visual assistance - all without breaking scrub. Congratulations to Adithya Shukla on an incredible project! Catch ORION in action → https://goo.gle/4ae2Dbi Read more in the blog → https://goo.gle/42Y2Sn5 Congrats to all the category winners and honorable mentions! 🏅The Live Agent: Bryen Param for drone-copilot → https://goo.gle/4aas9y7 🏅Creative Storyteller: Jeremiah Somoine for Sankofa → https://goo.gle/4wMq1X8 🏅UI Navigator: Enaiho Uwas Paul and Aman Kumar Sah for Moonwalk → https://goo.gle/4dAnGHo 🏅Best Multimodal Integration & User Experience: David Li for Wand → https://goo.gle/4a0FYzb 🏅Best Technical Execution & Agent Architecture: Matthew Keats for JohnKeats.AIhttps://goo.gle/49vmWRj 🏅Best Innovation & Thought Leadership: Yusuf Elnady for Rayan Memory → https://goo.gle/4uFFU0d ⭐ Honorable mention: Nikita Dongre and Omkar Dongre for NagarDrishti → https://goo.gle/4wSIxx6 ⭐ Honorable mention: Bassey John for Ekaette → https://goo.gle/4nWYa2C ⭐ Honorable mention: Michael Chang and 세준 김 for VibeCat → https://goo.gle/4dAPEml ⭐ Honorable mention: Sugam Palav, Nikhil Lohar, Siddhant Panday, and Vishal Parekh for Call My Parts → https://goo.gle/43x09Bc ⭐ Honorable mention: Faith Ogundimu for Relay → https://goo.gle/4o5AoBt A massive thank you to everyone who participated!

    • Against a dark background, a colorful four-pointed star icon sits above the text "Congratulations to the winners of the Gemini Live Agent Challenge," with "Gemini Live" in blue. The Google Cloud logo appears below the text, just above a vibrant, abstract wavy pattern of multi-colored lines along the bottom.
  • View organization page for Google for Developers

    4,058,235 followers

    🛠️ Build a long-running AI agent that pauses for days, survives restarts, and resumes without losing context using Google Agent Development Kit. Most agent tutorials teach you to build workflows that run for minutes. They show you how to chain a few API calls, stream a response, and close the connection. But real enterprise agents run for days or weeks. Think about employee onboarding: HR triggers a packet, the process pauses for 4 days while paperwork is signed, IT provisions accounts, and then it pauses another 3 days for hardware shipping. If your agent relies on in-memory sessions, it can fail the moment a container restarts or scales down. You cannot keep a container active and billing just to wait on human approvals. In our latest blog, Eric Dong and Shubham Saboo show you how to build durable, event-driven agents using the Agent Development Kit (ADK) on Agent Runtime. Here’s how the architecture handles multi-day timelines: 🔷 Durable state machines: Ground the agent in an explicit, deterministic state schema instead of relying on volatile chat history to remember its place. 🔷 Checkpoint-and-resume: Swap in-memory sessions for persistent SQLite or Cloud SQL. If a container restarts mid-workflow, the agent picks up exactly where it left off. 🔷 Event-driven sleep: Let the agent scale to zero during idle time. When a webhook fires (like a signed doc or shipping alert), the resume handler hydrates the session and wakes the agent. 🔷 Multi-agent delegation: Avoid prompt bloat by letting a coordinator hand off specialized tasks (like IT provisioning) to focused sub-agents. 🔷 Golden evaluations: Simulate days of waiting in seconds so you can test state machines in CI/CD without waiting weeks. Deploy to Agent Runtime with a single command to get built-in session persistence, auto-scaling, and Cloud Trace. Code + tutorial → https://goo.gle/4uGa8QL

    • This diagram illustrates an architectural shift from fragile "Stateless Agents" to durable "Long-Running Agents" capable of multi-day workflows. On the left, a stateless agent is depicted as forgetful and short-term, suffering from issues like context pollution and token cost explosions over 14 days because it can't survive idle time. On the right, a long-running agent uses persistent session storage and event-driven dormancy to manage a multi-step HR onboarding process, surviving restarts and pauses through durable memory schemas and multi-agent delegation.
  • Avoid common failure modes like brittle data formatting and unmonitored agent logic. Build a more reliable production architecture: 🧩 Use specialized sub-agents 🛡️ Force structure with Pydantic models 📈 Build a RAG pipeline 📊 Implement OpenTelemetry 🔌 Add circuit breakers Learn which approach fits your project best.

  • View organization page for Google for Developers

    4,058,235 followers

    The wait is over. 🍌 Nano Banana Pro and Nano Banana 2 are now GA on the Gemini API. This milestone signals a new era for developers and enterprises. As generative media moves into production-grade applications, the need for stability, consistency, and precision has never been higher. Which model is right for your workflow? 🍌 Nano Banana 2: Built on the Flash architecture, this is your enterprise workhorse. It delivers Pro-level fidelity with lightning-fast speed and cost-efficiency. 🍌 Nano Banana Pro: Designed for sophisticated creative control, this model excels at complex outputs and strict instruction following. Key production-ready features: • Subject consistency: Maintain resemblance for up to five characters across different scenes, a game-changer for storyboarding and brand narratives. • Advanced text rendering: Render crisp, accurate text directly into posters, marketing mockups, and greeting cards. • Real time info from the web: Improve factual accuracy in visual data, infographics, and diagrams. • Professional controls: Access 4K upscaling, custom aspect ratios, and advanced lighting/camera controls. We’ve already seen incredible innovation from our partners. Spline is streamlining 3D mesh generation, and krea.ai is enabling real-time complex image instructions. The era of vibe coding visual experiences is here. We can’t wait to see what you build. 🔗 Explore more in the blog: https://goo.gle/3RSaWmV

    • A grid of diverse thumbnail images featuring people, landscapes, products, and graphic designs is layered behind a dark, semi-transparent overlay. Large, central text announces "General Availability" in blue, followed by "Nano Banana Pro" and "Nano Banana 2" in prominent white lettering. This promotional graphic suggests the launch of new products or features related to the creative content shown in the background.

Affiliated pages

Similar pages