How I AI: My Live Test of Google I/O’s New AI Tools—From Gemini 3.5 Flash to Omni Video

Welcome back to How I AI. Today was the first day of Google I/O 2026, and the announcements came fast and furious. We saw new models, new developer tools, and a whole suite of creative features. It was a lot to take in, with so many product names that it’s hard to keep them all straight—Gemini, Antigravity, Flow, Stitch, Pomelli, and Omni, just to name a few.

In the spirit of How I AI, I couldn't just read the press releases. I had to get my hands dirty and see if the promise of these new consumer-grade and developer-focused features lives up to the keynote hype. I spent the afternoon putting these tools to the test, from the most technical coding assistants to the most fun and creative video generators.

This episode is my real-time reaction and a pragmatic guide to what builders and creators should actually pay attention to. We’ll look at what’s genuinely fast, what’s unfortunately broken, and where the real potential lies. I’ll walk you through several workflows I tried live on the show, sharing the exact prompts I used and the (sometimes surprising) results. Let’s separate the signal from the noise.

Developer Tools: The Good, The Fast, and The Not-Yet-Available

Google is clearly making a big push to get developers on its platform, with updates aimed at competing with tools we're already using. The theme seems to be catching up on features but trying to win on speed and multimodality, powered by their new models.

Workflow 1: Building an API with Antigravity 2.0 & Gemini 3.5 Flash

First, the foundation: the models. Google announced the Gemini 3.5 family, including Gemini 3.5 Flash, which they claim is their fastest and smartest coding model. The benchmarks position it as rivaling top models in intelligence while being significantly faster. The best place to test this is in their agentic IDE, Antigravity.

Antigravity 2.0 now includes features like Projects and Scheduled Tasks that feel familiar if you've used other AI coding assistants. I decided to give it a practical task for our own website, ChatPRD.

The Goal: We have an admin tool that uses video to generate blog posts, but it's a manual UI. I wanted to turn this into a programmatic API an agent could use.

Here was my step-by-step process:

1. I opened the Antigravity desktop app and loaded my website's project folder.
2. In the chat interface, which defaults to Gemini 3.5 Flash, I gave it a straightforward prompt:

We have a blog generator UI at, admin tools. I want to turn this into an API that an AI agent can use instead of a web UI for our team. Please build.

1. The agent immediately started analyzing the directory, asked for permissions (which I granted), and got to work.

The Outcome: It worked. It successfully created a new API endpoint with API key authentication to trigger the blog post generation. I can't say it felt four times faster, but it was competent and delivered a functional result without much fuss. The agent correctly identified the necessary files and created the logic I asked for.

I'm also really interested in the new slash commands they introduced, particularly /grill-me. It's designed to ask aggressive clarifying questions to get to the heart of a requirement, which sounds like a more direct version of Claude Code's polite questioning. That's one I'll be testing more.

Workflow 2: The Google Workspace Integration That Wasn't

Next, I turned to Google AI Studio, Google's low-code/no-code builder. The big announcement here was the ability to build apps connected to your Google Workspace data (Sheets, Gmail, Calendar, Drive). This is a huge use case for internal productivity tools, and I wanted to try it immediately.

The Goal: Create a simple personal app to manage my kids' weekend sporting events, pulling the data directly from my personal Google Calendar.

1. I went into Google AI Studio, ready to connect my account.
2. I wrote a clear prompt explaining what I wanted:

Make me an app to manage the next month of weekend events with my kids, in particular, our sporting events. They are all on my personal calendar in Google.

The Outcome: Nothing. The feature, despite being a highlight of the keynote, simply wasn't available in my account. I searched through settings and connectors and couldn't find any way to enable it. This was the first major instance of a theme for the day: a frustrating gap between what was announced and what was actually available to use. The vision is powerful, but for now, it's just a vision.

Creative Tools: Pushing Pixels, Videos, and Brand with AI

Beyond coding, Google is making a huge play in the creative space. They launched a confusing array of new tools and models, so I tested four distinct workflows to see where the magic is.

Workflow 3: Getting Creative (and Horrified) with Gemini Image Gen

The consumer-facing chat product, Google Gemini, got a redesign and an upgrade to its image generation capabilities, powered by a model called Nano Banana. I decided to give it a simple, vain task.

The Goal: Take a screenshot of me hosting the podcast and make it look better.

1. I dragged a screenshot of myself into the Gemini chat window.
2. I gave it a prompt:

upscale and beautify because I wanna be pretty. The image of this podcast host. Change the background to a professional podcasting studio.

The Outcome: A complete and utter failure, but a hilarious one. The result was, in my own words, "horrifying in every way possible." It was not my face. It was a photorealistic image of someone, but not me. It's a great reminder that while these models are getting better, image-to-image with specific faces remains a weak point.

Workflow 4: Bringing a Child's Drawing to Life with Omni

This is where things got exciting. Google announced a new video generation model called Omni, their competitor to models like Sora. It's built into Gemini and is supposed to be great at creating video from reference materials, leaning into Google's strength in multimodality.

The Goal: Animate a superhero drawing my kid made for me.

1. I took a screenshot of the drawing as I held it up to my camera.
2. I uploaded that image into Gemini's video creation tool.
3. I prompted it with a fun scenario:

animate this superhero, breaking a kid out of class to go. Have fun

The Outcome: A resounding success! The model generated a 10-second video that brought the hand-drawn character to life in a little animated story. The character was consistent, and the scene matched the prompt. This was the most impressive demo of the day for me. It shows the power of grounding video generation in a reference image and points to a future of truly creative, conversational video editing.

Workflow 5: The Failed Quest for a Digital Avatar in Flow

For more advanced video work, Google announced Google Flow, a tool focused on cinematic quality and character consistency. One of its buzziest features was the ability to create a digital avatar of yourself from a quick scan.

The Goal: Create an AI avatar of myself to use in future videos.

1. Inside Flow, I navigated to the 'Create an avatar' feature.
2. This generated a QR code that opened a web app on my phone (selfie.app.google).
3. I followed the on-screen instructions: read a series of numbers out loud and turn my head to the left and right for a 3D scan.

The Outcome: Another broken feature. After I completed the scan and it uploaded my data, the process just...failed. No avatar was created. I gave them my face, my voice, my identity—and got nothing in return. It was another example of a feature that wasn't ready for launch.

Workflow 6: Generating Brand and UI with Pomelli and Stitch

Finally, I checked out two tools for marketers and designers. Pomelli is for generating brand assets, and Stitch is for designing UIs.

The Goal: Use Pomelli to create a brand book from my website and Stitch to design an app screen.

• Pomelli: I gave it my website, ChatPRD. It successfully analyzed the site and generated a brand book with my colors and tagline. I then used its new feature to generate a website from that brand book. The result was... fine. It worked, but it wasn't inspiring.
• Stitch: Here, I prompted it to design me an app that covers the recent changes to Stitch in the style of the Google IO site. I pasted in the I/O site's URL. This worked beautifully. Stitch began streaming the design components directly onto the canvas, creating a mobile app screen in real-time. This felt like a great application of the fast Gemini Flash models in a creative context.

Final Thoughts: Promise and Problems

After a day of testing, my feelings about the Google I/O announcements are mixed. There is real power here. The core technology, especially the speed of Gemini 3.5 Flash and the multimodal video capabilities of Omni, is genuinely impressive. The success with the Omni video animation and the speedy UI generation in Stitch shows what's possible when the tech is applied well.

However, the user experience was littered with broken or unavailable features. The failed Workspace integration, the non-existent avatar in Flow, and the comically bad image generation in Gemini highlight a significant gap between keynote demos and product reality. It creates a sense of frustration and makes it hard to trust which announcements are ready for prime time. Not to mention the confusing mess of product names that makes it hard to even find the right tool.

My advice is to focus on the tools that work today. Developers should definitely give Anti-Gravity a try for well-scoped tasks where speed is a priority. And for creatives, the video generation inside Gemini is absolutely worth playing with. As for the rest, we'll have to wait and see if and when Google delivers on its promises. I can't wait to see what you build.

---

Brought to you by

• Magic Patterns—Prototypes that look like your product
• Thoughtspot—Build AI-powered analytics into your product

Episode Links & Resources

• Find me, Claire Vo, on my website or on X (formerly Twitter).
• Check out my agentic product management tool, ChatPRD.