AI Will Not Eat UI

I keep seeing takes on my feed about how AI will eat all of UI. That AI > UI. Maybe even all of software.

The basic premise: if AI can just do the task directly, why bother with a UI abstraction in the middle?

I think this is partly true, at least for the average case.

• Answering basic questions
• Scheduling a recurring meeting
• Redundant tasks will become much less dependent on UI

But the reality is, most tasks are complex. They require interactivity and reusability. Nobody wants to read a wall of dense text all day.

So here’s my take: AI will not eat UI. In fact, UI will become more important.

1. AI Coding for UI Will Be One of the Most Economically Important Ways People Build Things

Humans and agents will always be using UIs. In fact, UIs will become more important to use: better computer use, interfaces for humans post-code, legacy business software that needs maintaining.

There will be code written to build and maintain UIs.

As expectations and capabilities rise with AI progress, so will the need for new UIs and better versions of the old.

2. Most of the New UIs of the World Will Be Made Using Web Technology

Sub-bet: React is probably the last framework. Not literally the last framework ever, but the last major paradigm shift (JSX, hooks, etc).

3. Models Will Become (1) Incredibly Fast, Then (2) Incredibly Good at UI Tasks

Fast because code has verifiable rewards (tests pass, code runs), making it ideal for RL. Labs are scaling post-training on code, and coding tools are building on top. Good requires taste. Great people will get there eventually but slower.

4. Our Tools Will Go the Way of “Media Technology” Past

From daguerreotype to polaroid to camcorder to studio-level digital cameras back to iPhones. From low use to an explosion of capabilities and then a product for the average person.

Coding, unlike cameras, will progress OOM faster (5–20 years).

Once AI coding becomes commodified, the winners will be those with better taste and ease of use, not those with bleeding edge capability (1% better than the next best).

5. There Will Only Be Two Form Factors for Coding with UIs

First, low latency + low entropy (sub 100 ms), where the bottleneck is translating user intent. This is the latency threshold where interactions feel instant—anything slower breaks flow.

And long background tasks (big refactors, maintenance, scaffolding a well-spec’d feature), where the bottleneck is intelligence.

6. The Gap

There is no tool that is good at UI (yet).

No company is going all-in on AI coding for UI.