Open YouTube

Browsers and UIs are dead. Everything is chat

Syntaxautoenpublicupdated Jun 24, 2026, 14:22

Read in about 3 minutes instead of watching 18 minutes.

Chat Is Not the Only Interface

The speakers argue that claims like “the web is dead” or “chat is the final interface” are too simplistic; chat, websites, voice, and generated components all have different roles.
They reject the idea that every website should simply add a chat bar, because users often want agents to connect to sites and services rather than chat inside every site.
Cars are used as an example where a pure chat interface may be inappropriate, because drivers need fast, reliable controls while paying attention to the road.

Web MCP is described as a way for websites to expose their actions as tools that agents can use, while still letting humans use the normal website UI.
A grocery app example shows how actions like adding items, reordering, and marking items can be exposed to an agent through Web MCP.
The preferred model is flexible: sometimes users click, sometimes they prompt, and sometimes the system should infer or generate the right interface.

Visual interfaces remain important for tasks like calendars, shopping, social media, and browsing, where users want to scan, compare, explore, or consume content.
Inline chat components may work well for narrow workflows such as customer support refunds, but are unlikely to replace richer full-screen or specialized interfaces.
Shopping is presented as an experience many users enjoy because discovery, filters, recommendations, and sales are hard to reduce to a single prompt.

AI is most useful for annoying, repetitive work that people would rather delegate, such as categorizing expenses or adding calendar events from emails.
The speakers distinguish between automating busywork and preserving experiences people actually enjoy, such as browsing stores, markets, or art.
Connecting data across services is seen as a major strength of agentic systems, especially when that data is difficult to combine manually.

The speakers worry that chat-based access could put the open web behind a few powerful companies that control visibility, ranking, and monetization.
They note that companies may not want their brands and business flows reduced to APIs, especially where upsells and dark patterns are part of the existing experience.
They expect providers and businesses may make deals to insert sponsored results or dark patterns into agent-driven experiences.

Generated UI may land between fully hand-designed interfaces and fully AI-created HTML: companies could provide components while agents arrange them for the user’s needs.
The group questions whether an LLM or user can make a better interface than a company that deeply understands its own product and data.
Personalized generated interfaces may be useful for specific user goals, but they still depend on users or systems knowing what is actually needed.

Smart homes are presented as a strong use case because systems can use context such as cameras, humidity, light patterns, and recent events to suggest automations or dashboards.
The ideal smart home could learn routines, turn lights on automatically, or surface the right control button at the right moment.
The speakers remain cautious about whether AI can reliably learn patterns and make correct decisions, citing mixed experiences with navigation suggestions.

The speakers conclude that chat and agentic UI are new surface areas for applications, similar to mobile and voice, rather than replacements for the entire web.
They invite viewers to consider whether the web is truly dead or whether chat is simply another interface alongside existing ones.

Share your thoughts on whether the web is dead, whether everything will move into chat apps, or whether chat is just another interface.