Facebook Paper may have been ten years early. AI might finally make its idea work.

In 2014, Facebook shipped an app called Paper. It never went mainstream. Most people forgot it existed inside a year. A lot of us who work in UX never did.

Paper didn’t feel like a social app. It felt like a different idea of what a computer could be. Scrolling turned into exploration. Content had weight to it, like you could push it around with your hands. The controls got out of the way and let you touch the information directly.

Ten years later, I still think Paper got closer to a real natural user interface than almost anything shipped since. That should bother us. We’ve had a decade, and a wave of new hardware, and we still haven’t beaten what a small team was reaching for in 2014.

Here’s why I think we stalled. We had the wrong definition of “natural.”

We thought natural meant new inputs

For years, the industry treated natural user interfaces as an input problem. If we could get people to touch, talk, wave, or look instead of click, the interface would finally feel human.

So we got multi-touch. We got voice. We got motion tracking and gestures and eye tracking. The gear kept improving.

None of it changed the relationship.

A swipe was still a command. So was a voice request. So was a gesture. We traded the button for a wave of the hand and called it progress, but the human was still doing the translating. You still had to work out what the machine needed and then go perform it. The burden never moved. It just changed costumes.

I spent years in aviation Human Factors, designing for people who couldn’t afford to guess wrong. In that work, you learn fast that a “natural” control isn’t the one that looks slick in a demo. It’s the one that matches what the operator is already trying to do. Most of what got sold as NUI (Natural User Interface) failed that test. It looked natural and behaved like a command line with better manners.

The thing we were missing was intent

The problem was never the technology. It was the idea underneath it.

Computers understood actions. They never understood why you were taking them.

Old software works one way. The user decides which app to open, which menu to walk, which buttons to press, and in what order. Even a beautiful interface is a layer of translation sitting between what you want and what the machine can execute. You do that translating for free, every single time, and you’ve done it so long you stopped noticing the tax.

A real natural interface flips it. Instead of reading your commands, it reads your goal. You bring the intent. The system works out the execution.

That flip is the whole game. And it’s the part AI actually changes.

Why AI is different this time

For the first time, we have systems that can hold context well enough to matter. Not perfect. Not human. Good enough to close some of the gap between what you meant and what the machine has to do.

Say a product manager drops this on their team: “Let’s get everyone who worked on the checkout benchmark together next week and pull the top three issues.”

Old software makes you earn it. Open email, dig up the contributors, open the calendar, find a slot everyone can make, send the invite, collect the docs, build the agenda. Seven chores standing between one sentence and one meeting.

A system built around intent already knows who worked on the benchmark, which findings mattered, when people are free, and what got said last time. You state the goal. It handles the rest. And somewhere in there the interface stops being a thing you operate and starts being a thing you talk to.

Paper made content the interface. AI can go further.

Paper’s real move wasn’t the visual design people remember. It was the attempt to make content itself the interface, instead of a pile of chrome wrapped around content.

AI lets us take the next step. Context becomes the interface.

That sounds abstract, so let me ground it. Today we organize software around functions. Email lives in an email app. Calendar in a calendar app. Tasks, research, analytics, each in its own box. We built it that way because computers need structure.

People don’t think in apps. We think in outcomes. Nobody wakes up wanting to send an email. They wake up wanting a yes, or an answer, or a problem off their plate. The email is just the toll you pay to get there.

Apps exist for the machine’s convenience. Outcomes are what the human was after the whole time. AI gives us a real shot at deleting a lot of the structure we’ve been treating as permanent.

The end of navigation

We’ve spent decades getting good at helping people find things. Menus, search, dashboards, all of it is navigation. Even great UX still assumes the complexity is real and your job is to route people through it.

AI raises a stranger possibility. What if people mostly stop navigating?

Instead of “where do I find this,” the question becomes “what do I need to know.” Instead of “where’s last quarter’s report,” it’s “how did we do compared to last year.” The software shifts from storing and fetching to understanding and answering. Where the information lives stops mattering. What it means starts mattering more.

Context that doesn’t reset

One of the quiet failures of modern software is how much it forgets. Every app starts cold. So does every workflow. And every conversation makes you re-establish things the system three tabs over already knew.

A person carries context without trying. You know what project you’re on, why it matters, who’s involved, what happened yesterday and what a win looks like. Your tools mostly don’t.

A real natural interface holds that context across everything you do. Picture a system that already knows you’re building an exec presentation, that the room is senior leadership, that there are older benchmark studies on file, that last time the pushback was about sample size and that a couple of those people only care about business impact. Before you ask for anything, it’s already pulled the material you’re most likely to need. The tool stops reacting and starts working next to you.

From commands to conversation

The next interface may look like neither a command line nor a screen full of buttons. It may look like a conversation that keeps going.

This has nothing to do with chat being fashionable. People have always expressed what they want by talking. When we work together for real, we don’t bark commands at each other. We trade goals, questions, worries and half-formed observations.

A system worth building would handle lines like these:

“I’m worried leadership won’t buy these findings.”

“Something feels off about our conversion data.”

“Help me tell a sharper story with this.”

None of those are commands. They’re intent, stated the way humans state it. A natural system knows the difference and does something useful with it.

What the real NUI looks like

The best natural user interface might not involve gestures, voice, touch or a screen at all. It might just be an environment that keeps track of your context and responds to what you’re trying to do with almost no friction in between.

I’m not arguing to kill buttons. Or keyboards. Or even apps. Plenty of that stays. The target is the translation tax, the constant work of converting what you want into what the machine demands.

Paper hinted at a world where content mattered more than controls. AI points at one where context matters more than the interface itself. The graphical interface was about pushing windows around. Paper was about touching the content. The next era is about reaching past all of it and working straight with the outcome you were after.

That’s what natural user interface was always supposed to mean. For the first time, we might have the tools to pull it off.