Your AI agent shouldn’t answer in paragraphs. It should answer with UI. And that

By Chorouk Malmoum · October 31, 2026 · Curated by George's Blog

Your AI agent shouldn’t answer in paragraphs.

It should answer with UI.

And that now takes 2 lines of code!

Ask most agents to compare 3 pricing plans.

You get 4 paragraphs and a lot of scrolling.

Your user wanted a table.

There are 3 levels of agent UI:

1. Text: the agent writes, the user reads

2. Hand-built UI: you code a component for every tool

3. Generative UI: the agent builds the interface on the spot, from components you approved

Most teams are stuck on level 1 or 2.

Level 3 used to be a real project.

CopilotKit just released Catalogs and Surfaces, and it gets you to level 3 with basically two lines of code and one wrapper:

Here’s what you get:

→ 70 ready components (tables, charts, sliders etc)

→ Create/remove components

→ Surfaces: generate UI outside of the chat

→ Faster answers - transport via tool calls or markdown

→ Guardrails - the agent only renders from your catalog

It’s open source, ready for Google's A2UI 1.0 standard, and works with any AG-UI agent.

I left a link to their Docs in the first comment.

View the original post on LinkedIn