On Wednesday, March 18, someone described a billing dashboard to Stitch by voice. Got 5 screen variations in seconds. Dragged the one they liked into AI Studio, where Antigravity generated a Next.js app with Firestore and authentication. In 20 minutes they had something deployable.

This isn’t a future promise — it’s exactly what Google shipped this week. And it changes the type of conversation we’re going to have with clients.

These announcements also widen the gap between what’s possible and what’s production-ready. That gap is exactly where we live at IQ Source.

What Google built in two days

Stitch — design becomes a conversation

Stitch had been a Google Labs experiment for months. On March 18, they redesigned it entirely and named the category: vibe design.

The idea: describe what you need in natural language or by voice, and Stitch generates high-fidelity interfaces. These go beyond rough wireframes — actual screens with typography, colors, layout, and functional components.

What’s interesting isn’t the generation itself (several tools do something similar), but three specific things:

• Infinite canvas with voice: you can speak directly to the canvas. “Give me three sidebar menu options” or “switch the palette to dark tones” and it updates in real time. According to The Register, the agent can critique designs and interview you to understand what you need.
• DESIGN.md: Stitch generates a specification file that coding assistants can consume directly. It’s the design equivalent of a CLAUDE.md for engineering.
• SDK and MCP server: Stitch designs connect to Claude Code, Cursor, Gemini CLI, and other assistants — an open ecosystem, not a walled garden.

Beta pricing: 350 standard and 200 pro generations per month, free.

AI Studio + Antigravity — from prototype to working code

The next day, March 19, Google updated AI Studio with the Antigravity agent integrated. It’s an autonomous coding agent that understands the full project structure and makes multi-file changes with minimal human input.

According to the Firebase blog, the combination has already been used internally to build hundreds of thousands of applications.

What Antigravity does	What it doesn’t do
Generates apps in Next.js, React, Angular	Leaves out automated tests entirely
Detects when you need a DB and provisions Firestore	Missing monitoring and alerts
Adds Firebase Auth automatically	Lacks rate limiting or granular RBAC
Installs dependencies and external libraries	No built-in security reviews
Deploys to Firebase App Hosting	Requires manual integration for enterprise systems (SAP, Salesforce)

Firebase Studio shuts down — another forced migration

In parallel, Google announced the Firebase Studio sunset. Timeline: new workspace creation blocked June 22, 2026; full shutdown March 22, 2027.

Firebase Studio was less than a year old. Launched at Cloud Next in April 2025, already in its sunset period. As Frank van Puffelen, Firebase developer advocate, pointed out: the product will spend more than half its life in wind-down mode.

This aggressive sunset timeline is a recurring theme in Google’s ecosystem. I analyzed Google’s AI ecosystem dynamics and their implications for B2B companies a few weeks ago. The conclusion stands: Google consolidates aggressively, and companies that build on specific Google tools end up migrating every 12-18 months.

The full pipeline: from idea to deploy in minutes

When you put the three pieces together, the flow looks like this:

Step	Tool	Input	Output
Describe	Stitch	Voice, text, sketch	3-5 high-fidelity UI variations
Refine	Stitch canvas	Voice conversation	Final design + DESIGN.md
Build	AI Studio + Antigravity	DESIGN.md + instructions	Working app with Firebase (DB, auth, hosting)
Deploy	Firebase App Hosting	One click	Public URL with HTTPS

For internal prototypes and team tools, this flow is genuinely reliable. Someone with no design or coding experience can have something deployed in an hour.

The prototype works in the demo, but it’s far from viable in production.

Where it breaks for B2B companies

In our experience at IQ Source, the prototypes that cause the most damage are the ones that work perfectly in the demo. The director sees a polished app, assumes it’s “almost ready,” and expects a week or two of polish before client launch. What’s missing doesn’t show up in the demo.

What Google’s pipeline doesn’t cover — and that any real B2B application needs:

• Security: the generated code doesn’t go through an audit. No RBAC beyond basic Firebase Auth. No input sanitization or injection protection.
• Testing: Antigravity leaves out unit, integration, and end-to-end tests entirely. An app without tests is an app you can’t modify with confidence.
• Enterprise integrations: if your operations depend on SAP, Salesforce, HubSpot, a local ERP, or internal APIs, the pipeline doesn’t connect them. Firebase covers storage and auth. The rest is engineering.
• Observability: you won’t get structured logging, performance metrics, or built-in alerts. When something fails in production (and it will), you won’t know what happened or when.
• Regulatory compliance: if you handle financial, health, or personal data under privacy regulations, the generated code doesn’t meet audit or data retention requirements.

I covered the risks of vibe coding for enterprises in detail already, including dependency, licensing, and accumulated technical debt problems. All of it applies here.

Stitch’s MCP changes the equation

The most relevant part of the announcement for engineering teams is the SDK and MCP server.

MCP (Model Context Protocol) is the standard that lets AI tools share context. With Stitch’s MCP, designs flow directly to Claude Code, Cursor, or Gemini CLI. Instead of a static export, it provides live context the coding assistant can query while building.

The DESIGN.md file Stitch generates serves a similar function to CLAUDE.md for engineering: structured documentation the agent consumes to make implementation decisions. This is context engineering applied to design.

The pattern I see working for serious engineering teams:

1. Explore with Stitch: use vibe design for fast UI iteration. Instead of meetings debating “what if the table goes here,” generate 5 variations and decide.
2. DESIGN.md as specification: the file Stitch generates becomes the source of truth for design, consumable by both humans and agents.
3. Real engineering for production: the team writes the code (with assistants), not the automated pipeline. With tests, with review, with continuous integration.

As I wrote about the SDLC collapse, the design and development phases are merging. Stitch is a concrete example of that merger. But that merger doesn’t eliminate the technical work — it transforms it.

When to use this pipeline and when not to

Use case	Does the pipeline work?	Recommendation
Internal team dashboard	Yes — ideal	Stitch → AI Studio → direct deploy
Quick MVP validation	Yes — with caution	Full pipeline for the demo, real engineering if validated
Client-facing SaaS	Design phase only	Stitch for exploration, own code for production
Regulated industry (fintech, health)	Design only	No production with unaudited generated code
ERP/CRM integrations	No	Pipeline doesn’t connect to legacy enterprise systems
High-availability app	No	You need your own infra, monitoring, failover

What we’re doing with this at IQ Source

This week we started testing Stitch with clients — as an accelerator for the exploration phase, not a replacement for the full design process.

The concrete value for our B2B clients: what used to be 2-3 weeks of back-and-forth between mockups and feedback compresses to 2-3 hour sessions with Stitch. The client describes what they need, we see variations in real time, and we leave with a DESIGN.md that the team can use as a specification.

The MCP integration is the part we’re most interested in. We already use Claude Code for development. Having designs flow directly into the coding agent’s context eliminates the manual translation that normally gets lost between design and development.

We make sure every client understands a hard truth: the prototype is not the product. Google’s pipeline gives you the visible 20% in 5% of the time. The remaining 80% (security, tests, integrations, observability, compliance) still requires engineering.

Send us a one-paragraph project description. We’ll run it through the pipeline and return a working prototype plus an honest assessment of what you need to reach production.

Send project description