Last Wednesday, one of my engineers pinged me: “Have you seen the Stitch update? It just generated a better onboarding flow than our designer did in two days.”

I was skeptical. I’d tried the original Google Stitch last year and dismissed it as a glorified wireframe generator. But the March 18 update — what people are calling Stitch 2.0 — is a fundamentally different product. So I blocked off three afternoons and ran it through our actual design pipeline.

The verdict? It’s the first AI design tool that made me rethink a process. But it’s not replacing anyone’s job yet.

What Actually Changed in the March 2026 Update

Google Labs rebuilt Stitch from the ground up into what they call an “AI-native infinite canvas.” In practice, this means:

  • Multi-modal input: You can feed it text prompts, voice descriptions, screenshots of competitor apps, hand-drawn sketches, or even code snippets. It synthesizes all of these into editable UI concepts.
  • Design agent: An AI agent that reasons across your entire project history — not just individual prompts. It remembers your style preferences and iterates on previous outputs.
  • Code export: Direct export to React, Flutter, and SwiftUI. Not just HTML/CSS like before.
  • Free tier: Completely free through Google Labs. No waitlist as of this writing.

The canvas feels less like Figma and more like a Miro board that can build things. You drag ideas around, connect them, and Stitch infers relationships between screens.

Test 1: Mobile Onboarding Flow (Score: 8.5/10)

I described our app’s onboarding flow in plain English: “Four-step onboarding for a fitness tracking app. Step 1: set your fitness goals. Step 2: set notification preferences. Step 3: pick a subscription tier with a comparison table. Step 4: confirmation with confetti animation.”

Stitch generated a complete four-screen flow in about 14 seconds. The visual hierarchy was solid — better than I expected. The subscription comparison table used proper card layouts with highlighted “recommended” badges.

What impressed me:

  • Color consistency across all four screens without me specifying a palette
  • The comparison table was genuinely usable, not just a placeholder grid
  • Voice follow-ups worked: “Make the team selection a horizontal scroll instead of a grid” — it got it right instantly

What didn’t work:

  • The confetti animation was referenced in the design but obviously can’t be previewed
  • Exporting to React produced 380 lines of inline styles. No component library integration, no design tokens
  • Custom illustrations were generic — acceptable for prototyping, terrible for production

Test 2: Admin Dashboard (Score: 4/10)

This is where Stitch fell apart. I prompted: “Admin dashboard with a sidebar nav, three data tables with sort/filter, a chart panel with line and bar charts, and a notification drawer.”

The output looked like a template from 2019. The data tables had no real interactivity affordances. The sidebar was comically wide. The chart section used placeholder images instead of actual chart components.

When I tried to refine via voice — “Make the sidebar collapsible and add a breadcrumb nav” — it regenerated the entire layout instead of modifying the existing one. Three attempts, three completely different designs. No continuity.

The core problem: Stitch is noticeably better at mobile-first, content-light designs. Anything with data density, nested states, or complex interaction patterns produces mediocre results. If you’re building admin tools, dashboard-heavy apps, or anything with more than two data tables on a screen — Stitch isn’t ready.

Test 3: Landing Page Redesign (Score: 7/10)

I uploaded screenshots of three competitor landing pages and said: “Create a landing page that combines the hero section style of screenshot 1 with the social proof layout of screenshot 2, but use a darker color scheme.”

This is where the multi-modal input shines. Stitch correctly extracted design patterns from the screenshots and recombined them. The hero section had the right proportions, the testimonial cards used a layout similar to screenshot 2, and the dark theme was applied consistently.

Downsides:

  • It copied the competitor’s section ordering almost exactly — not much creative reinterpretation
  • The responsive breakpoints in the exported code were wrong. The mobile version stacked elements awkwardly
  • No SEO meta structure in the HTML export. Had to add that manually

How It Compares to Our Current Workflow

We currently use Figma → developer handoff with a fairly manual process. Here’s my honest comparison:

Stitch wins at:

  • Speed from zero to first mockup (minutes vs. hours)
  • Exploring multiple design directions quickly
  • Mobile-first designs for simple flows
  • Stakeholder demos when you need “something visual” fast

Figma still wins at:

  • Design systems and component libraries
  • Complex, data-heavy interfaces
  • Collaborative editing with designers (Stitch’s collaboration is barebones)
  • Production-ready design tokens and handoff specs

Neither wins at:

  • Bridging the actual gap between “looks good in mockup” and “works in production.” Both still require significant engineering effort to go from export to shippable code.

The Real Question: Who Is This Actually For?

After a week with it, I think Stitch 2.0 is most useful for:

  1. Solo developers who need decent-looking UIs without hiring a designer. It’s dramatically better than Bootstrap templates.
  2. PMs and tech leads who want to mock up ideas before involving the design team. I’ve already used it twice to communicate a feature concept to my team.
  3. Early-stage startups that need to validate ideas fast and can tolerate design imperfections.

It’s not for teams with established design systems, complex enterprise UIs, or anyone who needs pixel-perfect output. At least not yet.

Three Things Google Should Fix

  1. Export quality is embarrassing. Inline styles, no component library support, no TypeScript. The React export looks like AI-generated code from 2023. If they want developers to actually use this output, it needs Tailwind/shadcn support at minimum.

  2. No design system import. I can’t feed it our existing Figma design tokens and say “use our color palette and typography.” Every project starts from scratch.

  3. The “agent” loses context too easily. On complex refinements, it regenerates instead of editing. This turns iterative design into a slot machine — you keep pulling and hoping for a better result.

Bottom Line

Google Stitch 2.0 is the best free AI design tool available right now. For mobile mockups and landing pages, it’s genuinely useful — I saved roughly 3-4 hours on two quick prototypes this week. But calling it a Figma replacement is like calling a microwave a replacement for a kitchen. It does one specific thing well and everything else poorly.

The most interesting thing about it isn’t the current output quality — it’s the trajectory. If the March update is this much better than the original, the version shipping in six months could be legitimately threatening to design tool incumbents.

I’ll be using it for quick prototypes and stakeholder demos. I won’t be canceling our Figma subscription.

Try it yourself: stitch.withgoogle.com

Related reading: If you’re interested in AI design tools, I also covered Figma’s Code to Canvas — a different approach to the same problem from the other direction.