Using AI still feels like work.

Not slow work.
Not hard work.
Interruptive work.

That interruption is so ingrained in our routine that we hardly notice it anymore. But it’s based on an old belief about computers: If you need help, you go to the machine. That made sense when software was confined to separate applications with distinct tasks. It feels less logical now that computers can grasp what you’re trying to do in plain language yet still require you to step away from your work to ask for assistance.

Press enter or click to view image in full size

The part that never improved

Most folks tend to use AI in a pretty similar way.

You hit pause on your writing. You fire up your assistant. You type out a version of the idea that was already brewing in your mind. Then you wait. You read what it generated. You copy it. You paste it. And then you scramble to remember what you were thinking before you got sidetracked.

On a technical level, that process is quite impressive.

But as a way of thinking, it feels like a tax.

Have you ever gone back to a document after asking AI something and spent a good ten seconds trying to recall what you were about to say?

That’s the real cost.

Not just those ten seconds.

It’s the disruption of your train of thought.

Every time you switch contexts, your brain is asked to do some unpaid work. You have to reload where you left off. Reconnect with your intention. Hunt down that almost-formed sentence. Get back into the groove of the problem. Those seconds might seem trivial on a timer, but they’re costly when it comes to maintaining your flow.

And once that flow is disrupted, it’s not just speed that you lose. You also lose those half-formed ideas that only come to life when you’re deeply engaged in something.

It’s a peculiar setup: the most capable collaborator you’ve ever had, yet every time you need it, you have to step away from your desk, walk to another room, rehash the entire situation, and then walk back before you can pick up where you left off. We wouldn’t design an office like that. Yet, somehow, we’ve come to accept it as the norm for software.

The real bottleneck in AI isn’t its intelligence.

It’s the interaction.

Press enter or click to view image in full size

We optimized the wrong thing

For decades, every major advancement in computing has followed a familiar pattern.

Faster processors.
Faster networks.
Faster storage.
Better software.

We kept pushing to make machines more powerful, but we rarely changed how quickly we could turn a thought into action. AI shakes things up. Unlike past advancements, it’s not just about hardware; it’s all about how we interact with it.

Picture this: you have a brilliant idea while drafting an email, fixing a bug, or brainstorming a new product. But before the computer can lend a hand, you have to convert that idea into something it can understand — a tab, a prompt, a chat thread, or a clipboard. That’s where the real friction lies.

One little interruption? No big deal. But twenty? That’s a day gone by. And when you hit hundreds, it becomes your entire workflow. Eventually, you might not even realize you’ve trained yourself to navigate through all these detours. Everyone seems to think the future is all about smarter models, better reasoning, and improved answers.

But I believe the real breakthrough lies elsewhere: it’s about creating smarter interactions, reducing interruptions, and enhancing our thinking. We’ve built computers around software, but now it’s time to build software around humans.

For the past forty years, we’ve adapted to computers. We’ve learned the ins and outs of shortcuts, menus, ribbons, tabs, windows, dialogs, and terminals. We’ve become fluent in navigating interfaces because that’s what it took to use these machines. Now, computers can finally grasp human language almost as effortlessly as we speak it. So why are we still making humans adapt to computer interfaces? That’s the true shift that AI enables — and it’s something we haven’t fully embraced yet. We’ve added intelligence to the mix, but we’ve left the way we interact with it unchanged.

In 1985, the interface was the bridge.

In 2025, AI often becomes another stop on the way to the interface.

Tomorrow shouldn’t add another layer. It should remove one.

The next leap in computing won’t come from the smartest model.

It will come from the interface that no longer interrupts the way we think.

What’s actually broken

The models were never the whole story.

What broke first was the relationship between thought and action. We built astonishing intelligence, then wrapped it in the same interaction pattern we’ve used for decades: stop, switch, explain, wait, return.

We spent the last few years asking:

How do we make AI smarter?

The more important question is simpler.

How do we make interacting with it disappear?

Because the future of computing isn’t about making machines think more like humans.

It’s about making computers work the way humans already think.

We didn’t need better AI.

We needed a better way to think with it.