Gemini and beyond: tools and models

Year: 2025 · ▶ Watch on YouTube

Debi Cabrera (Senior Developer Advocate)

Switch language → zh

Segments (3)

  • 00:05:520 · Introduction — Debi Cabrera
    • The speaker introduces the topic of empowering developers to use their favorite tools and models on Google Cloud.
  • 00:33:800 · Demo: Gemini in Your Favorite IDEs — Debi Cabrera
    • Demonstrating how Gemini can be used for AI-assisted coding in three different IDEs: Windsurf, Cursor, and IntelliJ with GitHub Copilot.
  • 04:47:340 · Demo: Your Favorite Models on Google Cloud — Debi Cabrera
    • Showcasing Vertex AI Model Garden for accessing various models like Llama 3.3 and demonstrating how to easily switch to a Google Cloud-hosted model (Claude 3.7) in application code.

Products Announced (3)

  • 04:22:980 · Gemini 2.5 Model Family in IDEs (Available)
    • Integration with Windsurf, Cursor, IntelliJ (via GitHub Copilot) · Support for AI-assisted coding, refactoring, and test generation · Also available in VS Code, Tabnine, Cognition, and Aider
    • Available out of the box in partner IDEs.
  • 05:18:030 · Vertex AI Model Garden (Updated)
    • Access to Google models (Gemini, Imagen 3, etc.) · Access to partner models (Llama 3.3, Gemma 3, Anthropic, Mistral) · Deploy models from registries like Hugging Face
    • Models can be used as a managed service or deployed to a user’s endpoint.
  • 05:28:600 · Llama 3.3 API Service (Preview)
    • Access Meta’s Llama 3.3 models as a fully managed Vertex AI service · Test and use the model without managing infrastructure · Offered as a pay-per-hour service
    • Available in public preview at no cost for testing; pay-per-hour for deployment.

Competitor Mentions / Comparisons (3)

  • 05:24:910 · vs Meta — Mentioned as a partner providing the Llama 3.3 model in Vertex AI Model Garden.
  • 05:25:510 · vs Anthropic — Mentioned as a partner whose models (like Claude 3.7) are available on Vertex AI.
  • 05:26:880 · vs Mistral AI — Mentioned as a partner providing models in Vertex AI Model Garden.

Commitments (1)

  • 07:08:000 (Ongoing) — We’re striving to meet developers where you are.

Demos (5)

  • 01:11:800 ✓ · Windsurf IDE with Gemini 2.5 Pro — Debi Cabrera
    • The demoer used a natural language prompt (‘Build me a Java microservice app…’) to have Gemini scaffold an entire project structure, including a pom.xml file and multiple service files.
  • 02:49:380 ✓ · Cursor IDE with Gemini 2.0 Flash — Debi Cabrera
    • The demoer used an inline command (‘add input validation’) to have Gemini automatically add validation logic and error handling to an existing Java controller file.
  • 03:48:470 ✓ · IntelliJ with GitHub Copilot and Gemini 2.0 Flash — Debi Cabrera
    • The demoer used a Spanish prompt (‘agregar pruebas unitarias’) in the GitHub Copilot chat to generate unit tests for the existing Java codebase, demonstrating multilingual understanding.
  • 05:28:600 ✓ · Testing Llama 3.3 in Vertex AI Model Garden — Debi Cabrera
    • The demoer used the Model Garden UI to send a prompt to the Llama 3.3 API service and received a detailed, structured response directly in the browser.
  • 06:35:680 ✓ · Switching to a Vertex AI-hosted Model — Debi Cabrera
    • In VS Code, the demoer made two small code changes in a Python script to switch from calling the Anthropic Claude model directly to calling the same model hosted on Vertex AI, simplifying infrastructure and improving latency.

Notable Quotes (4)

  • 00:50:800 — Debi Cabrera:

    Have you heard of vibe coding? People are doing a lot of it.

  • 01:48:900 — Debi Cabrera:

    I’m gonna take the escalator.

  • 04:02:600 — Debi Cabrera:

    Oh, muy bien, Gemini!

  • 05:06:900 — Debi Cabrera:

    So, behold, Vertex AI’s Model Garden.

Visual Signals

On-screen (9)

  • 00:05:520 · Debi Cabrera, Senior Developer Advocate, Google Cloud
    • Introduces the speaker and her role.
  • 00:23:700 · Gemini and beyond: Your favorite tools and models
    • Title slide for the presentation segment.
  • 00:34:300 · Windsurf IDE UI with 'Write with Cascade' and 'Gemini 2.5 Pro'
    • Introduces the first IDE and the specific Gemini model being used.
  • 02:13:200 · Cursor IDE UI with Java code
    • Introduces the second IDE being demonstrated.
  • 03:24:000 · IntelliJ IDE UI with GitHub Copilot chat and 'Gemini 2.0 Flash'
    • Introduces the third IDE and model combination.
  • 04:50:000 · Vertex AI Model Garden UI showing Gemini, Imagen 3, and WeatherNext models.
    • Visually introduces the Model Garden platform and key Google models.
  • 05:24:910 · Partner logos in Model Garden: ANTHROPIC, META, HUGGING FACE, MISTRAL AI
    • Shows the breadth of third-party model providers available on the platform.
  • 05:29:200 · Llama 3.3 API Service page in Google Cloud Console
    • Shows the specific landing page for using Meta’s model as a managed service.
  • 06:11:800 · Python code in VS Code showing an import for 'anthropicvertex'
    • Sets up the final demo of running a partner model on Vertex AI.

Stage (2)

  • 00:05:520 · Debi Cabrera walks on stage to applause.
  • 01:48:000 · In a staged bit, Debi Cabrera pretends to take an ‘escalator’ off one side of the stage while another speaker takes an ‘elevator’ up, facilitating a transition between demo machines.

Visual demos (5)

  • 00:34:300 · Windsurf IDE
    • A dark-themed IDE with a file explorer on the left and a chat-based ‘Cascade’ interface on the right. The user types a prompt, and the AI generates code and file structures.
  • 02:13:200 · Cursor IDE
    • A dark-themed IDE showing a Java file. An inline editing prompt appears, and the AI’s suggested code changes are highlighted in green directly within the editor before being accepted.
  • 03:24:000 · IntelliJ with GitHub Copilot
    • A dark-themed IntelliJ IDE with a Java file open on the left and the GitHub Copilot chat panel on the right. The user types a prompt in the chat, and the AI generates a new file with unit test code.
  • 04:50:000 · Vertex AI Model Garden
    • The Google Cloud Console UI for Model Garden, showing a catalog of models with cards. The user navigates to the Llama 3.3 page and uses a simple prompt box to test the model’s output.
  • 06:11:800 · Visual Studio Code
    • A dark-themed VS Code editor showing a Python file. The speaker highlights specific lines of code that are changed to switch from a direct API call to using the Vertex AI SDK.

Key Topics

AI-assisted coding · Developer Experience (DevEx) · IDE Integration · Gemini · Vertex AI · Model Garden · Multi-model support · Third-party models · Llama 3.3 · Claude 3.7 · Agentic coding · Code generation · Microservices · Java · Python

Takeaways

  • Google is heavily investing in integrating its Gemini models directly into the most popular developer IDEs, including Windsurf, Cursor, IntelliJ, and VS Code.
  • The focus is on developer choice, allowing teams to use their preferred tools and models within the Google Cloud ecosystem.
  • Vertex AI Model Garden acts as a central hub for discovering, testing, and deploying both first-party (Google) and third-party (Meta, Anthropic, Mistral) AI models.
  • Developers can easily run third-party models like Llama and Claude on Google Cloud infrastructure with minimal code changes, benefiting from co-location, lower latency, and simplified security.
  • AI coding assistants powered by Gemini can handle complex, multi-step tasks like scaffolding entire applications, refactoring code with specific instructions, and generating unit tests from natural language prompts.
  • Gemini models demonstrate strong multilingual capabilities, as shown by understanding and executing a request given in Spanish.
  • Google provides managed API services for popular open models, abstracting away the complexity of infrastructure management for developers.