Google's A2UI 1.0: When AI Assistants Answer With a Booking Form, Not a Wall of Text

StarkLabX ·

Last week Google released A2UI 1.0, a stable open protocol that lets AI agents answer with real screens — forms, charts and booking pickers — instead of a wall of text. Here is what that means for your website, and the work we would start with at StarkLabX.

If you have asked ChatGPT, Gemini or a website chatbot to book a table, compare three products or pick a delivery slot, you already know the weak point. The assistant can talk. What you usually get back is a wall of text: a numbered list, a polite “which one would you like?”, and another round of messages.

That is a poor way to compare options or confirm a booking. On 11 November 2026, Google released A2UI 1.0, the first stable version of its open protocol for generative UI. In plain language: an AI agent can now answer with a real screen — a form, a chart, a date picker, a confirm button — and your app draws that screen with its own components.

We are StarkLabX, a web and app development studio in Batticaloa. This is not a reason to rip out your website tomorrow. It is a clear signal that customer journeys are moving beyond text chat, and that businesses with structured catalogues, bookings and clean web apps will be in a better position than businesses that only exist as a Facebook page or a chatbot.

What Google actually released

A2UI is an open-source generative UI framework. The agent describes what to show. The host app renders it with native components from a catalogue you control, on the web, Android or iOS. Google’s announcement is explicit about the split: the model does not get to invent raw, executable interface code and drop it into your product. It composes from components you have registered, in your design system.

That matters for two reasons. The screen can look like your brand, not like a generic chat bubble. And sensitive values can stay on the server, bound to the UI, so they do not have to be poured through the model.

Google says teams at Atlassian, Alation, Bloomberg and ByteDance are already building with it. Inside Google, Gemini Notebook already uses interactive cards to confirm options before generating a deck, and organisations can register their own agents that respond with A2UI. Gemini itself still supports A2UI 0.9.1 today, with work underway to move Gemini Enterprise up to 1.0.

The official site is a2ui.org. The project lives at github.com/a2ui-project/a2ui.

What is new in version 1.0

Version 1.0 is the first stable protocol. The changes that matter for a business product are practical, not just a version bump.

  • Bidirectional function calls. A flight or reservation picker can check live availability on the spot, instead of showing a stale list and hoping it is still free.
  • Accessibility and extensions. The protocol adds accessibility support and a way to carry extra metadata.
  • Official renderers. Open-source libraries draw A2UI screens with React, Angular and Lit on the web, Flutter on mobile, web and desktop, and SwiftUI on iOS. Jetpack Compose for Android is at 0.9.1, with 1.0 coming soon.
  • Agent SDKs. There are SDKs for Python, TypeScript and Dart, to help an agent generate, translate and correct A2UI messages.
  • It can appear inside other assistants. The web and Flutter renderers also work inside MCP Apps, so the same surface can show up in AI assistants that host MCP Apps, such as Claude and Gemini.

Google’s own examples are easy to picture. Ask an agent to move a dinner reservation, and it can answer with one view that lines up your calendar, open tables and a ride — and nothing is sent until you confirm. That is a different product from a chatbot that pastes a paragraph and asks you to type “option 2”.

Why this matters for Sri Lankan businesses

Plenty of customers here already start with a message: WhatsApp, Instagram, a website form, or an AI assistant. A hotel in Ella, a restaurant in Batticaloa, a clinic in Colombo or a shop that takes islandwide orders all get the same kind of question. What is available on Friday? What does this cost? Can I book for four people?

If the only place your prices, rooms, menu or time slots live is in someone’s head, or in a WhatsApp broadcast, an assistant has nothing reliable to show. If they live in a real website or app — a catalogue, a booking engine, an inventory that stays up to date — then a future agent can present that information as a screen the customer can actually use.

This is the same gap we already see without any new protocol. A brochure site and a phone number cannot confirm a booking. A text chatbot that guesses from an old PDF cannot either. A2UI does not fix that on its own. It gives developers a safer way to turn a good backend into an interactive answer, in your own app or, later, inside a third-party assistant.

You do not need to implement A2UI this week to benefit from the idea. You do need the foundations it assumes: structured data, a booking or catalogue system, and a website that is more than a digital business card.

What we would start with at StarkLabX

When a client asks, “Should we add AI to the site?”, we usually start one layer down.

First, can a customer already complete the job on your website? Book a room, pick a product, choose a slot, pay, or send a complete enquiry, on a phone, without a staff member typing back. If they cannot, a smarter chat box will not save the experience.

Second, is the information trustworthy? Hours, prices, stock, delivery areas and available dates have to be current. An assistant that shows last month’s price is worse than no assistant.

Third, is the interface yours? A2UI’s design is built around a catalogue of components you approve. That only helps if you have a real product UI to begin with — forms, cards, calendars, confirm buttons — not a pile of plugin shortcodes or a theme you cannot change.

That is the work we already do: custom websites and web apps, booking systems, catalogues and checkouts, built so you own the code. Making those systems “agent-ready” is a natural next step, not a separate gimmick. We will not pretend every small business needs an agent in Gemini tomorrow. We will say that the businesses getting ready now are the ones with a solid site and a clean source of truth.

A practical checklist

  • Give customers a real website or app, not only a Facebook page or a WhatsApp number on a poster.
  • Put products, services, rooms or time slots in a structured catalogue or booking system, with prices and availability you can update.
  • Make the main actions work on a phone: enquire, book, add to cart, confirm.
  • Keep hours, prices and stock consistent on the site, on Google Business Profile, and in any chat tool you already use.
  • If you add a chatbot, connect it to the same live data. Do not let it invent availability.
  • Prefer interfaces you control — your forms, your cards, your confirm step — over raw AI-generated HTML.
  • If you later expose an agent to a third-party assistant, register only the components and actions you are willing to stand behind, and never send a booking or payment until the customer confirms.
  • Watch the official docs at a2ui.org as Gemini, Claude and other hosts adopt the stable 1.0 protocol. Do not depend on unofficial “AI UI” plugins that promise a one-click upgrade.

Common mistakes we still see

  • Treating a generic website chatbot as the product, instead of fixing the booking or catalogue behind it.
  • Leaving the real price list and availability only in WhatsApp chats or Instagram captions.
  • Publishing a brochure site, then hoping an AI layer will “make it interactive”.
  • Letting a model draft raw HTML or take an action before the customer has confirmed.
  • Ignoring accessibility, which A2UI 1.0 now includes in the protocol for a reason.
  • Waiting for a perfect standard, and doing nothing about a slow, outdated or unbookable website in the meantime.

Frequently asked questions

What is A2UI, in one sentence?

A2UI is an open protocol, now stable at version 1.0, that lets an AI agent describe an interactive screen — a form, a chart, a booking picker — and lets your app draw that screen with its own native components.

Is this the same as adding a chatbot to my website?

No. A chatbot answers in text. A2UI is about answering with a real user interface, using components you have approved, including inside other AI assistants that support MCP Apps. A chatbot without a booking system, catalogue or live data is still just a conversation.

Do Sri Lankan businesses need to adopt A2UI today?

Not as a rush job. Renderers and host support are still rolling out — for example, Gemini supports A2UI 0.9.1 today, and Jetpack Compose 1.0 is still on the way. The useful move today is to have a solid, structured website or app that customers can already use to book or buy. That is what an agent would need later.

Will this replace my website?

No. The protocol assumes there is a host app with a design system and a source of truth. Your website or app is still the place customers confirm a booking, see your brand, and trust that the price is right. Assistants become another door into that same system, not a replacement for it.

What should I do if I only have a Facebook page or a brochure site?

Start with the customer job you care about most: a bookable service, a catalogue with real prices, or a clear enquiry flow that works on a phone. Once that exists and stays up to date, you are in a much better place to add agents or richer AI interfaces later.

Want a website an assistant can book with, not just talk about?

StarkLabX builds custom websites, booking systems and catalogues that customers can use today, and that are ready for this kind of agent interface tomorrow. If you want a second pair of eyes on your current site, we are happy to talk it through.

Talk to us: starklabx.com · hello@starklabx.com · WhatsApp +94 76 881 6881


Sources

These are the official pages we used while writing. We have not added unofficial adoption numbers or invented statistics.