Kill the interface

AI is not a feature. It’s a new interaction model.

A colleague told me something a few weeks ago that stuck with me more than it probably should have.

“AI is going to bring a big change in tech in the next 3-4 months.”

Not “in the next few years.” Not “eventually.” Next 3-4 months. I didn’t fully agree with the timeline, but I couldn’t stop thinking about the confidence behind it. Because he wasn’t talking about a new model release or a benchmark score. He was talking about a shift in how people expect to work. And that’s a very different claim.

It made me want to write down what I actually think is happening here, before I build anything else under this section.

The AIFication of everything

You’ve seen the joke by now. Slap “+ AI” onto literally anything, a candle, a toothbrush, a fridge, and suddenly it’s a product category. It’s funny because it’s true, and it’s true because we’ve seen this exact pattern before.

We had physical work. Then came machines. And the fear back then was simple: if machines do the work, people go unemployed. That didn’t happen. Machines didn’t erase work, they erased tasks, and humans found new layers of work sitting right above the old ones. Operating the machine. Maintaining it. Designing what it should do next.

AI is the same pattern playing out one layer up. It’s not erasing work, it’s erasing tasks. And the joke about “+AI” selling more isn’t really about AI being magic. It’s the market’s way of saying: we already sense the layer is moving, we just don’t know exactly where the new work will land yet. So everyone’s slapping AI onto everything, half as a bet, half as a hedge.

That’s layer one. Here’s layer two, and it’s the one I actually care about as a builder.

Thinking becomes the only skill worth protecting

As humans, we’ve always tried to eliminate redundancy. Not because we’re lazy, but because attention is the scarcest thing we have. Every tool we’ve built, from the wheel to the spreadsheet, is really just an attempt to free up consciousness for something better than repetition.

AI is the most aggressive version of that attempt yet. It doesn’t just remove physical redundancy, it starts removing cognitive redundancy: the parts of thinking that were never really thinking, just retrieval, formatting, navigating.

Which leaves an uncomfortable but clarifying question: if the tool can do retrieval, formatting, and navigation for us, what’s actually left for us to do?

My answer: judgment. Framing the right problem. Deciding what’s worth doing at all. That’s the layer AI hasn’t touched, and it’s the layer that decides whether everything below it is even pointed in the right direction.

Once you accept that, you stop asking “what should I automate” and start asking a more interesting question: what is the interaction supposed to look like now that thinking is the scarce resource, not clicking?

That question is what this whole section is built on.

---

The real shift

Everyone in product right now is asking:

“How can we add AI?”

Almost no one is asking:

“Should this interaction even exist anymore?”

The biggest shift AI brings isn’t better autocomplete. It’s a change in the relationship between users and software.

Traditional software

Traditional products ask the user to learn the interface.

User → Clicks → Navigation → Forms → Feature → Output

The burden sits on the user. You learn the menu structure. You remember where things live. You adapt to the software’s model of the world.

AI products

AI reverses that relationship.

Intent → Conversation → Context → Reasoning → Action → Result

The burden shifts to the product. The software adapts to your model of the world, not the other way around.

Example 1: Google Docs

Traditional:

Find menu → Insert → Table → Rows → Columns

AI:

“Create a comparison table for the three pricing plans.”

The interface disappears. You stop thinking in menus and start thinking in outcomes.

Example 2: Canva

Old: Choose template → Resize → Move text → Adjust spacing → Export

AI: “Create an Instagram carousel explaining Newton’s First Law.”

Again: less software, more intent.

The mistake companies are making

Most companies are doing this:

Existing Product + Chat Window = "AI Product"

It isn’t. That’s feature inflation: bolting a new input method onto an old workflow and calling it transformation. Real AI adoption doesn’t add a box to type into. It changes the shape of the workflow itself.

The new builder question

Instead of asking:

“Which AI feature should we build?”

Whoever builds the product should be asking:

“Which user workflow no longer needs an interface?”

That’s a much harder question, and a much more useful one.

Implications for builders

Future builders won’t just prioritize features. They’ll redesign workflows.

Instead of thinking in:

• dashboards
• filters
• menus

They’ll need to think in:

• memory
• context
• delegation
• verification
• trust
• human oversight

These aren’t UI components. They’re design primitives for a completely different kind of product.

The AWARE framework

A simple lens I keep coming back to when I evaluate whether an “AI feature” is actually doing anything:

• A: Automate repetitive work
• W: Work from intent, not navigation
• A: Augment decisions, don’t replace them
• R: Retain user control
• E: Explain important outputs

If a feature can’t hold up against these five, it’s probably a chat window bolted onto an old workflow, not a rethink of the interaction itself.

Closing

Machines didn’t end work. They moved it up a layer, and gave us more of it. AI is doing the same thing, faster, and one layer closer to thinking itself.

AI won’t replace every interface. But every interface now has to justify why it still exists.

That’s the lens I’m building the rest of this section through.

first written on notion