A request for a dashboard comparing current language models exposed a major limitation: Stitch generated a polished interface but populated it with stale or invented model information. Even supplying a source website did not make it reliably extract current data from that page.

Visual generation does not guarantee factual accuracy
The generated model data appeared to come from training data
Explicit requests for current internet data were ineffective
Providing a source website did not reliably ground the dashboard

“So it didn't look like it grounded from the internet because we can see this is all from 2024, which is probably when the sort…”
Matt Wolf · 07:00

“Yeah, it doesn't appear to be able to pull information directly from the web to build out your dashboard, but we did get a dashboard…”
Matt Wolf · 07:30

#data grounding#llms#hallucinations#dashboard design

Hot Take· 1

Hot Take02:00

Landing Pages That Took a Week Can Now Take a Minute

Stitch could radically shorten the landing-page production cycle for marketers who previously depended on page builders and manual design work. It can also generate alternative layouts and headlines quickly, making creative experimentation far less expensive.

Compress landing-page production from days to minutes
Reduce dependence on specialist design resources
Generate modern designs through conversation
Create split-test variants without rebuilding pages manually

“We would spend a week on a landing page, right? And now it's like I can have that same kind of thing done, but probably…”
Matt Wolf · 02:30

“Now it's like I literally can speak with my voice and say, create me a split test version with a couple different headlines, and it'll…”
Matt Wolf · 02:30

#landing pages#marketing#split testing#workflow automation

Explainer· 3

Explainer01:00

How Google Stitch Turns Prompts Into Website Designs

Google Stitch applies the vibe-coding concept to visual design, allowing users to generate websites and web-app interfaces through natural-language prompts. It presents the results on a Figma-like canvas where users can review and iterate on generated screens.

Generate interface concepts from natural-language prompts
Create websites, landing pages, and web-app screens
Review generated designs on a visual canvas
Use the tool without traditional design expertise

“Well, apparently this is that, but for designing, for designing websites, for designing web apps, things like that.”
Matt Wolf · 01:30

“But now you can actually just prompt your own designs into existence.”
Matt Wolf · 02:00

#google stitch#vibe design#web design#generative ai

Explainer08:30

Why Stitch and AI Studio Serve Different Roles

Stitch is strongest as the visual design phase, while Google AI Studio provides the environment for developing functionality and additional pages. Exporting a Stitch design transfers assets and generated files into a simplified coding workspace.

Use Stitch primarily for visual design
Export generated assets and code to AI Studio
Treat Stitch's HTML preview as a basic prototype
Develop functionality in the coding environment

“But Stitch is clearly designed to be your design phase.”
Matt Wolf · 09:30

“But you want to throw it into something like AI Studio to like start dialing in the functionality.”
Matt Wolf · 09:30

#google ai studio#prototyping#code generation#design handoff

Explainer12:00

Publishing Requires a Google Cloud Project

AI Studio can publish the generated application to a public URL while keeping its chat history and code private. Publishing requires selecting a Google Cloud project, and a custom domain can be connected later through Google Cloud.

Select a Google Cloud project before publishing
Publish the application to a public URL
Keep chat history and code private
Connect a custom domain later through Google Cloud

“Your chat history and code will stay private. Your app will be accessible via a public URL.”
Matt Wolf · 12:00

“And then when you're ready to connect it to a domain name, you would do that from inside of Google Cloud.”
Matt Wolf · 12:30

#google cloud#deployment#publishing#custom domains

Story· 1

Story03:00

Stitch Generated Three Distinct Future Tools Redesigns

Matt tested Stitch by asking it to redesign Future Tools. After analyzing the existing site, Stitch inferred its colors and typography, replicated the current aesthetic, and then produced three substantially different design directions.

Analyze an existing website's visual identity
Infer colors, fonts, and styling
Begin with a close visual replication
Generate multiple layouts and aesthetic directions

“It actually figured out like the color schemes and the fonts and everything just by looking at the website.”
Matt Wolf · 03:00

“But each one is like a completely different layout, different color scheme, different design, and all really good modern professional-looking designs.”
Matt Wolf · 03:30

#future tools#design variations#style system#website redesign

Tool· 1

Tool09:30

Stitch Can Generate a Predictive Click Heat Map

The design environment generated a heat map estimating where visitors would click on the proposed interface. This brings a capability that traditionally required separate analytics or optimization software directly into the design workflow.

Estimate likely click locations
Evaluate attention before publishing
Use Google's behavioral data advantage as a possible signal
Reduce reliance on separate heat-map tools

“So it's generating a predictive heat map. People pay good money to get predictive heat maps.”
Matt Wolf · 09:30

“It created a heat map of where it thinks people will most likely click.”
Matt Wolf · 10:00

#heat maps#conversion optimization#analytics#user behavior

Takeaway· 3

Takeaway10:00

Stitch Prototypes Work, but Design Consistency Can Drift

Stitch successfully turned a generated screen into a coded prototype, but adding another screen caused some visual changes. The result remained impressive while demonstrating that automated multi-screen prototyping still requires review for aesthetic consistency.

Convert designs into working prototypes
Generate additional screens with natural-language prompts
Expect occasional visual drift between screens
Review generated screens before treating them as final

“Although, interestingly, so it kind of tweaked the design a little bit.”
Matt Wolf · 10:30

“It's still pretty impressive that you can just like make a design like this, right click, export it over to AI Studio, and have it…”
Matt Wolf · 10:30

#prototyping#design consistency#ai limitations#web apps

Takeaway11:00

Build the First Screen in Stitch, Then Expand in AI Studio

Matt recommends using Stitch to establish the initial design and moving to AI Studio for subsequent pages. Because AI Studio works from the exported scaffolding, it can preserve the established layout while focusing on implementation rather than repeatedly redesigning the product.

Establish the initial visual direction in Stitch
Export the selected design to AI Studio
Add pages on top of the existing scaffolding
Keep layout and aesthetics consistent during expansion

“I would probably just build the initial design over in Stitch and then build out the rest here.”
Matt Wolf · 11:00

“It will keep all of the existing sort of layout and aesthetics and everything, and just build the next page for you.”
Matt Wolf · 12:00

#workflow#google stitch#google ai studio#page development

Takeaway13:00

Stitch Appears Free and Earned a Rare Strong Endorsement

At the time of testing, Matt could not identify a direct cost for using Stitch, though future or undiscovered usage limits may apply. Despite its factual-grounding and consistency limitations, he regarded it as one of the most impressive recent AI products he had tried.

No obvious usage cost was identified
Undisclosed or future limits may still apply
The tool impressed an experienced AI reviewer
Current strengths outweigh notable prototype limitations

“I don't think there's any cost involved, at least not yet.”
Matt Wolf · 13:00

“I mean, this is one of the more impressive things that I've played with lately.”
Matt Wolf · 13:00

#pricing#product review#google stitch#ai tools

Frameworks in this episode

Generated-Data Grounding Check

Multi-Variation Design Exploration Loop

Stitch-to-AI Studio Design-to-Build Pipeline

Insights & moments

⚡Myth Buster· 1

Stitch Did Not Ground Its Dashboard in Current Web Data

◆Hot Take· 1