Vibe coding to production: AI Studio evolution

Year: 2026 · ▶ Watch on YouTube

Sam Witteveen (Host) · Logan Kilpatrick (Head of AI Studio)

Switch language → zh

Segments (12)

  • 00:00:00 · Introduction and First Impressions — Sam Witteveen, Logan Kilpatrick
    • The speakers introduce themselves and share their excitement about being live at Google Cloud Next, noting the dominant theme of AI agents.
  • 00:01:29 · The Evolution of AI Studio — Sam Witteveen, Logan Kilpatrick
    • Logan discusses the journey of AI Studio from a simple prototyping tool (‘MakerSuite’) to a full-fledged production environment.
  • 00:02:44 · AI Studio’s ‘Build’ Tab and Vibe Coding — Sam Witteveen, Logan Kilpatrick
    • Logan explains the ‘Build’ tab, a ‘vibe coding’ experience that allows users to generate working apps from prompts, including new features like design previews and ‘Tap Tap Tab’.
  • 00:05:21 · The Rapid Pace of AI and Model Improvement — Sam Witteveen, Logan Kilpatrick
    • The speakers discuss how quickly AI capabilities are advancing, making things possible now that were not just a year ago, and how this raises the ambition for what can be built.
  • 00:08:45 · Voice, Multimodality, and ‘Yap to App’ — Sam Witteveen, Logan Kilpatrick
    • They cover the power of multimodal models, specifically voice input, and the ‘Yap to App’ feature that turns spoken ideas into functional app prompts.
  • 00:10:02 · AI Studio on Mobile and On-Device Models — Sam Witteveen, Logan Kilpatrick
    • Logan hints at upcoming mobile support for AI Studio and the potential of using on-device models like Gemma for local processing.
  • 00:11:22 · Agentic Engineering and the Future of Development — Sam Witteveen, Logan Kilpatrick
    • The conversation shifts to ‘agentic engineering’ (vibe coding rebranded), the partnership model between AI and human engineers, and how it improves code quality.
  • 00:17:32 · The Next Billion Users and the Creator Economy for Software — Sam Witteveen, Logan Kilpatrick
    • Logan compares the democratization of software creation to YouTube’s impact on video creation, empowering a new generation of builders who don’t need to be traditional coders.
  • 00:26:42 · The GenMedia Portfolio: Imagen, Veo, Lyria — Sam Witteveen, Logan Kilpatrick
    • They discuss the consolidation and power of Google’s generative media models, built on the foundational Gemini architecture.
  • 00:30:31 · The Future of Coding and Robotics — Sam Witteveen, Logan Kilpatrick
    • Logan identifies coding and robotics as the next major frontiers for AI, with robotics being about 18 months behind where coding is today.
  • 00:35:03 · DeepResearch Agent and Long-Running Agents — Sam Witteveen, Logan Kilpatrick
    • They discuss the newly updated DeepResearch agent and the future of long-running agents that can operate for days or weeks without human intervention.
  • 00:42:38 · Closing Remarks and Call for Feedback — Sam Witteveen, Logan Kilpatrick
    • Logan encourages the community to provide feedback on AI Studio to help guide its development.

Products Announced (4)

  • 00:08:08 · AI Studio 'Edit Mode' (Rolling out this week)
    • Allows targeted edits to generated apps · Draw on a preview to specify changes · Select and regenerate specific elements like images
    • Included with AI Studio
  • 00:33:34 · DeepResearch Agent (Updated this week)
    • First agent available in the Gemini API · Enables long-running research tasks · New ‘Max’ version for more rigorous and deeper research
    • Available now in the Gemini API
  • 00:04:36 · AI Studio 'Tap Tap Tab' feature (Newly Landed)
    • Uses Flash model for autocomplete · Generatively extends a user’s app idea prompt · Helps users articulate more complex and ambitious app concepts
    • Available now in AI Studio Build tab
  • 00:10:25 · AI Studio Mobile (Coming Soon)
    • Brings AI Studio experience to mobile devices · Aimed at reaching the ‘next gen developer audience’ · Will potentially integrate on-device models like Gemma
    • More to share soon

Commitments (2)

  • 00:28:46 (Ongoing) — Google is pushing the rock up the hill on coding models, with a huge amount of investment happening.
  • 00:41:30 (Ongoing) — The AI Studio team is committed to deploying technology to ensure it benefits humanity and to gather user feedback to improve the product.

Demos (2)

  • 00:23:36 ✓ · Refrigerator Repair with Multimodal AI — Sam Witteveen (describing a personal experience)
    • Sam described using a live multimodal AI to diagnose a broken refrigerator by showing it the warning light, where the AI identified the model, the problem (filter change), the filter location, and helped order a new one.
  • 00:24:26 ✓ · Coffee Machine Helper with Astra — Logan Kilpatrick (describing a past demo)
    • Logan described using Project Astra (precursor to Gemini Live) to get step-by-step instructions on how to use a complex coffee machine by showing it to the AI via a camera.

Notable Quotes (6)

  • 00:00:25 — Logan Kilpatrick:

    It feels like the era of agents is upon us.

  • 00:00:56 — Logan Kilpatrick:

    It’s delivering, which I think is really, really exciting. And I think we’re at sort of still inning or chapter number one of that actually playing out.

  • 00:15:38 — Logan Kilpatrick:

    It’s not that the model… like I can’t be like, ‘Oh, well the model can’t do it, so it’s fine.’ Like now the onus is on me to be like, I really could build this.

  • 00:19:03 — Logan Kilpatrick:

    Now everyone is a builder. Everyone can build. And the technology is enabling that to happen.

  • 00:40:13 — Logan Kilpatrick:

    As this total addressable market of the number of people making software increases, it increases the demand for developers because there will be a stopping point of how far you can go if you don’t understand all the detail of how this technology works.

  • 00:36:53 — Logan Kilpatrick:

    I think that will be… maybe the only reason why what you’re saying is not true is like, Gmail just becomes an agent.

Visual Signals

On-screen (3)

  • 00:00:01 · Google Cloud Next '26 logo and branding
    • Establishes the event context for the interview.
  • 00:00:19 · Google Cloud Next '26 logo on the microphone boom arm
    • Reinforces the event branding in the physical set.
  • 00:05:01 · Lower third with 'Google Cloud Next' text
    • A blue animated lower third appears, branding the segment.

Stage (3)

  • 00:00:00 · The interview begins in a podcast-style booth set up on the main floor of the Google Cloud Next conference, with attendees visible in the background.
  • 00:04:55 · The camera switches to a wide shot showing both speakers at the desk, with the ‘Google Cloud Next’ branding prominent on the desk front.
  • 00:43:14 · The speakers wave goodbye to the camera, concluding the interview.

Key Topics

AI Agents · AI Studio · Vibe Coding · Agentic Engineering · Gemini · Multimodal AI · Developer Experience · Software Development · On-device AI · Robotics · Generative AI · Google Cloud Next · Democratization of Technology · Text-to-Speech · Long-running Agents

Takeaways

  • The ‘era of agents’ is here, moving beyond hype to real, delivering products that can perform complex, long-running tasks.
  • AI Studio is evolving from a prototyping tool into a full ‘vibe coding’ environment where users can build and deploy production-ready apps directly from prompts.
  • New AI Studio features like ‘Tap Tap Tab’ and ‘Yap to App’ are designed to help users, even non-coders, articulate and build more ambitious software ideas.
  • The rapid improvement in AI models, particularly multimodal capabilities like live voice and video, is unlocking new use cases like real-time, personalized tutoring and troubleshooting.
  • The democratization of software creation is a major trend; just as YouTube created a ‘creator economy’ for video, new AI tools are creating a ‘builder economy’ for software.
  • This new wave of builders will increase, not decrease, the demand for traditional developers, who will be needed to handle more complex tasks and partner with AI-empowered creators.
  • The future of AI development will focus on consolidating various specialized models (image, video, music, speech) into more unified, powerful agentic systems.
  • Robotics is seen as the next major frontier for AI, currently about 12-18 months behind the curve of coding models, but poised for significant breakthroughs.