“Computer, show me the crew manifest.” No login. No hamburger menu. Just a question and an answer. Star Trek’s been doing this since 1966. Sixty years, and we still haven’t built it.

The Window Cleaner’s Invoice Screen

In 2022 I was building an invoicing app for a local window cleaner. Django backend, HTMX and Alpine.js on the front. Straightforward stuff. The hard part wasn’t the code. It was figuring out how this guy’s brain worked.

He didn’t think in forms and dropdowns. He thought in “I cleaned these windows, now I need to get paid.” Every screen I built was me guessing at the gap between what he wanted to do and what the software needed him to click. I wrote a lot of boilerplate to generate those screens, and most of it was there to manage complexity that shouldn’t have existed.

Building custom software, you’re not really building features. You’re trying to reconstruct how someone else thinks, and then hoping you didn’t get it wrong. Pendo’s research says 80% of features in the average software product are rarely or never used. Eighty percent. That’s a lot of interface nobody asked for.

Mercury OS

Around the same time, I stumbled on Mercury OS. Jason Yuan, an ex-Apple interaction designer, had put together this concept for a speculative OS. No apps, no folders. Modules that show up based on what you’re trying to do, generated from natural language. He called them “browsers capable of generating UI on-demand.”

The design language was so minimal it was almost invisible. You see what you need, nothing else.

I didn’t buy the AI parts of it at the time. But the core idea, that an interface should be transparent enough to disappear, stuck with me. I was already a heavy Raycast user, doing most things through a command palette instead of clicking around in apps. Mercury OS was that same impulse taken to its logical extreme.

What if there were no interfaces for anything? Not practical, I know. But the question wouldn’t leave me alone.

Every App Is Its Own Language

To use Netflix you learn Netflix. Where the categories live, how search works, what the algorithm is doing to you. Switch to Hulu and you learn Hulu. Same deal with email. Gmail works one way, Fastmail works another, and the task hasn’t changed. You’re still reading and sending messages.

Universal patterns exist. Nobody reinvents the scroll bar. But every new app still has its own learning curve. Menus, settings, little workflows you have to internalize before you can do the thing you came to do.

Interfaces aren’t bad. They have to exist. But they tend to rot. Features pile up, options get nested inside other options, stuff gets buried in hamburger menus three levels deep. What started as a clean screen becomes an archaeological dig.

Computer, Do This

I came to Star Trek late. Started watching it in the last few years. What struck me wasn’t the warp drive or the phasers. It was the computer.

“Computer, what is the current position of the Romulan vessel?” And it just... answers. You get a screen when you need one, and that’s it. Kirk was doing this in 1966. Picard refined it in 1987. This has been on television for sixty years.

We’re in 2026 and we still don’t have it. Yes, there are YouTube videos of people building “Jarvis” with voice APIs and some Python scripts. Those prove you can yell at your computer. They don’t prove much else. Saying “open Spotify” with your voice is still Spotify’s interface. You skipped one click.

The Star Trek computer doesn’t open apps. You tell it what you need and it figures out the rest. Completely different thing.

The Brains Are Arriving

Something’s different now, though. Not a big announcement or a breakthrough—more like a handful of tools that work the way you’d want them to.

OpenClaw lets you talk to an AI through Telegram or WhatsApp, whatever chat app you already have open. You message it like you’d message a coworker: clear my inbox, reschedule that meeting. And it does it. The interface is whatever you were already using.

Mario Zechner’s pi is a coding agent that sits in your terminal. You describe what you want, it reads your files, writes code, runs commands. Four tools total. That’s the whole thing. I’ve been using it for a few days and it feels more like working with someone than using something.

I’m not declaring victory. Siri promised this ten years ago and delivered a weather widget. But these tools feel different. They don’t shove a bunch of UI between you and the thing you’re trying to do.

I dictated the initial thoughts for this post into a voice transcription, fed them to a coding agent, and it organized them into what you’re reading now. Not “personal log, star date 47634.44.” But closer than anything I’ve used before. I wrote about this from a more analytical angle elsewhere. This is the personal version.

A website can’t be a chat box. A medical records system shouldn’t be a command palette. Interfaces aren’t going anywhere. But I keep noticing the question has changed. It used to be about what the interface should look like. Now I catch myself asking how much of it needs to exist.

ai, tools, opinion, web