A designer friend told me a couple of years ago that the AI revolution was going to be a design revolution. Someone had to figure out the human-agent interface layer. New metaphors, new primitives, the next Macintosh. She wanted her team at the forefront.
Two years later, the most-used AI products in the world look like this: a text box. Sometimes the text box has a microphone. Sometimes a paperclip. Mostly it’s just a text box.
ChatGPT: text box. Claude: text box. Gemini: text box. Perplexity: text box with a vibe. Cursor and Windsurf: code editor with a text box bolted on. Devin: a chat window watching itself work. Operator and other computer-use agents: a chat window watching a browser. Replit Agent: chat. Voice mode: no interface at all.
There is no design revolution happening in AI. There is an *undesign* revolution. The interface is collapsing into a prompt, and increasingly into nothing at all.
The Tweet
A post by signulll sharpened my thinking on this, and the whole thing is worth reading. His core argument: the modern computer probably needs to be reinvented from scratch. Not improved. Not given a chatbot sidebar. Rebuilt from first principles the way the iPhone rebuilt the pocket computer.
The current setup, with its windows, cursors, apps, files, and folders, assumes a human staring at a screen, manually translating intent into interface actions. That setup made sense, he writes, when “the human was the runtime.”
In an AI-native world, the human stops being the runtime. The agent is. And nearly everything we built, including files, folders, apps, desktops, browsers, and the operating system itself, was scaffolding for the human runtime. Much of it stops making sense.
His framing of computer-use agents is the part that lodged in my head. They’re useful, but they’re transitional. We’ve trained models to drive software that was built for human hands and human eyes. Clever, sure, but a bit absurd when you sit with it. He likens it to engineering a robotic hand that can grip a doorknob, when the better question is whether a door needs a knob at all.
He poses a list of first-principles questions worth sitting with:
- What is a file when the system already understands the context you’d be using it for?
- What is an app when intent can route itself to the right capability?
- What is a desktop when work gets decomposed, executed, monitored, and summarized by agents instead of clicked through by hands?
- What is a browser when the agent does the searching, comparing, buying, and remembering?
- What is an operating system when the primary user isn’t a single person, but a person plus a swarm of delegated intelligences, or no person at all?
He summarizes the shift in a sentence: the old computer’s job was organizing information; the new computer’s job is organizing agency.
Where Are All the UI/UX Designers?
This is the question that started this post for me. A lot of designers I know expected the AI era to be a design renaissance. New categories everywhere. The agent dashboard, the trust-and-permissions layer, the memory canvas, the approval queue, the multi-agent monitor. Designers were going to be back in the driver’s seat the way they were in the early iPhone era.
What actually happened: the dominant AI products replaced graphical interfaces with conversation. Conversation is a content surface, not a design surface. A skilled designer can make a text box prettier, but they can’t redesign the shape of one. The shape is the shape.
The actual frontier in AI design today isn’t visual; it’s conversational, behavioral, and trust-based. What does the agent say when it isn’t sure? When does it ask permission? How does it surface its reasoning? How does it handle interruption, correction, undo? How does it tell you it’s about to do something irreversible? These are real design problems, but they’re closer to game design or screenwriting or product policy than they are to traditional UI/UX. Most of them get owned by product managers, prompt engineers, and model researchers, not designers.
The handful of teams trying to design new agent surfaces, like Rabbit’s R1, Humane’s pin, and the various “AI OS” startups, have mostly produced demos that didn’t survive contact with users. The market keeps voting for the text box.
Do We Even Need Computers Anymore?
Here’s the harder version of the question. If the agent is the runtime, and the agent works in language and tool calls, what is a personal computer actually *for*?
A reasonable answer: computers will increasingly be for computers. Servers running models, GPUs running training, agents calling other agents over APIs. The human-facing surface shrinks down to whatever’s needed to issue intent and verify results, which on most days is a phone, a microphone, and a screen for the occasional bit of output you want to look at.
This puts legacy interfaces in a strange spot. The macOS Finder, the Windows desktop, the Chrome tab strip, Excel, Photoshop. These are deeply optimized for the old runtime. They aren’t going to disappear, because there’s still a long tail of work where the human really is in the loop. But they stop being where most of the work happens. They become specialty tools, the way a CAD program is a specialty tool today. The center of gravity moves to the prompt.
And the prompt, eventually, moves off the screen. Voice mode is already most of the way there. Glasses are coming. Ambient agents that act on your calendar and inbox without you opening anything are already shipping. The endpoint isn’t a redesigned computer. It’s no computer.
The Hacky Middle
We’re not at the endpoint. We’re in what signulll calls the hacky middle stage. Sidebars, copilots, agents clicking through legacy UI, automation layers sitting on top of forty-year-old metaphors. Most “AI products” today are really just chatbots glued to existing software, or existing software with a chatbot glued to it.
The new computer, if it ever ships, will probably treat memory, context, identity, permissions, tools, agents, and interfaces as native primitives. First-class citizens of the OS rather than features bolted onto an OS designed for someone clicking a mouse in 1984. Desktop, mobile, browser, apps, files, folders all deserve another first-principles look.
Whether designers get to lead that look, or whether it gets handed to whoever is closest to the model, is the open question. My bet is that the people who define the next computer will not be the people who defined the last one. The skill set has changed. The runtime has changed. And the interface, increasingly, isn’t an interface at all.