The blank page is the most honest interface in the world. It can become anything. A novel, a business plan, a resignation letter.

And you're stuck.

Not because you can't write. Because nothing is telling you where to start. Every possibility is open, which means no possibility is suggested. Too much option is no option.

Now look at the interface we keep being told is the future: a single chat input. "How can I help you today?" One field, infinitely capable, just describe what you want.

It's the same blank sheet. Infinite possibility, zero direction.

For a general tool, a ChatGPT, a search box, that's fine. You arrived with a question and you type it. But the moment a tool is for doing work, building a page, running a campaign, editing a design, that empty input quietly transfers all the thinking onto you. What's possible here? What should I ask for? What does this thing even do?

Power isn't the bottleneck anymore. Direction is. And a blank input gives you none.

This post is about the opposite of blank. About interfaces that do the thinking before you arrive, and about a simple rule for when to give people a button and when to give them the chat.

The rule: button or chat

Chat versus buttons gets treated like a war. One side says everything becomes conversation. The other says chat is a fad. Both are wrong, because it was never either-or. It's a sorting problem, and here's the rule I use to sort it.

If an action fits on one screen, at one level, behind one absolutely clear icon and a label of two words max, it's a button. Publish. Undo. Duplicate. Align left. Don't make me describe it. Let me hit it.

If the action needs a sentence, if I'd have to explain what I want with context and conditions, it's chat. "Take last quarter's pricing page, keep the layout, swap the testimonials for enterprise logos, and make the tone more formal." There is no icon for that. There shouldn't be.

There's also a tell for when you've sorted wrong. If a task is buried three dropdowns deep behind settings nobody can find, it was never a good button. It was a sentence wearing a button costume. Free it and give it to the chat. And the reverse: if people keep typing "make the line height 1.5" into a chat and waiting for a model to think, that was always a button. A slider. A field. Show it.

Buttons for the instant and obvious. Chat for the specific and explained. Simple. But this only solves half the problem, because someone still has to decide which buttons you see. And that decision is the entire game.

Eggs or pancakes

Here's how I explain it at home.

If I ask my wife "what do you want for breakfast?", I've handed her a blank sheet. Every dish on earth is technically on the table. She has to scan the fridge in her head, weigh options, decide. I've given her work disguised as a question.

If I ask "eggs or pancakes?", I did the work. I already checked what we have, what she likes, what fits a Tuesday morning. Two good options, one second to choose, zero pain. She gets breakfast faster, and thinking about it cost her nothing.

That's what curation is. Not limiting people. Doing their first three steps of thinking for them, so they only make the choice that actually matters.

And the options aren't universal. If I'm asking my daughter, it's a different two options, shown differently, probably with less negotiation about chocolate. Same question, different person, different menu. Knowing which two options to offer is knowing your user. That's the skill.

Now put that on software. A tool that opens with a blank input asks "what do you want for breakfast?" A tool that opens with the right two or three suggestions, for you, today, asks "eggs or pancakes?" One of these gets things done.

The right 30%

So which options do you show? It depends who's standing in front of the tool.

A marketer building a landing page thinks in campaigns, audiences, conversions. The tool should talk that language: patterns they already know, controls named the way they name things. A designer needs different tools entirely: type, spacing, components, the full cockpit. Same engine underneath. A different 30% on the surface.

Because that's the real ratio. Nobody uses a whole product. Everybody uses their slice of it. Surface the right 30% for a role and 80% of that person's work is one reach away. Surface everything and you're back to the blank sheet with extra steps.

It matters most for the person who isn't the expert. Say you're not a marketer, but today you have to do marketing. A blank input is the worst thing you can hand this person. They don't know the vocabulary, they don't know what's possible, they're stuck before they start. Suggestions, guidance, three reasonable starting points with sane defaults: that pushes them into motion.

Starting somewhere always beats starting from scratch. Every expert interface should have a beginner's on-ramp built from suggestions, not from a cursor blinking at you.

The car already solved personalization

At this point someone always says: but people love personalization, don't lock things down.

Right. And a car already solved this.

Every car on earth: steering wheel, pedals, indicators. Same place, same behavior. Those are the essentials, and nobody gets to move them. That's why you can drive any car on sight.

But the seat? Fully yours. Mirrors, height, colors, the whole cabin mood. Adjust everything. Some cars even remember it: press one button and everything returns to your setup.

Essentials standardized, preferences personal. Switch to another car and the essentials are exactly where your hands expect, while your adjustments come along or take a minute to redo. Nothing important had to be relearned.

That's the contract for interfaces too. The core patterns, where things live and how actions behave, stay standard so skill transfers. On top of that, personalize freely: your views, your shortcuts, your defaults. Personalization is the seat. It's never the steering wheel.

The tool that remembers yesterday

There's one more kind of blank sheet, and it's sneaky: the tool that forgets you.

"Hey, what are we doing today?" Sounds friendly. It's a blank input with a smile. Now I have to remember where I left off, reload the whole context, find the thread.

Compare: "How about finishing that script from yesterday? Or maybe one of these three things." Now the tool did the remembering. Maybe I'd even forgotten that task, and here it is, one tap from resuming. And if the suggestions are wrong? Fine, I say what I want instead. The suggestion cost me nothing, but it saved me the reload.

Continuity is curation across time. Yesterday's context is the eggs-or-pancakes of today's session.

Where personalization ends

One boundary worth naming: other people.

Set up your own workspace however you like. That's the seat. But the moment work is shared, common patterns take over. If every teammate's setup is a foreign country, you spend the meeting learning each other's UI instead of doing the work. Shared time should go to the work, not to translation.

Personal freedom at the edges. Common grammar at the center. Same contract as the car.

Someone designs the defaults

So, generative UI. It's real, it's coming, and it's good news, as long as we're honest about what it needs. Left alone, it hands everyone the blank sheet. Curated, it hands everyone eggs or pancakes.

But curation doesn't happen by itself. Someone decides which two options a marketer sees. Which 30% surfaces for which role. What's a button and what's a sentence. Which essentials never move, and where personalization runs free.

That someone is a designer. Not designing the screens anymore; the screens get generated. Designing the choices. The patterns, the defaults, the rules the generation runs on.

Blank is the most expensive interface. The most valuable thing you can ship is the thinking that means nobody has to face it.

I'm unpacking how to actually build that system, piece by piece, in the next post.