The Next Interface, and the Toll Booth Behind It

Every leap in computing has been a leap toward ease. The history of how we talk to our machines is essentially a history of removing friction between human intent and machine action.

In the beginning there was the command line: a blinking cursor that demanded you know the exact incantation. Computers belonged to scientists, engineers, and hobbyists, and a steep learning curve marked the pre-GUI era. You didn't click anything — you recalled the right text command and typed it precisely, or nothing happened.

Then came the graphical user interface. Xerox PARC pioneered the mouse-driven, windowed desktop in the 1970s, but it was Apple's Macintosh in 1984 that brought the GUI to the masses. The shift was profound: instead of recalling obscure commands, users could now recognise options by seeing them on screen, which dramatically lowered the cognitive load of using a computer. Point and click. You no longer needed to understand the machine to use it.

The next great simplification was touch. With Apple's iPhone in 2007, and later the iPad, the pointer disappeared entirely. You reached out and touched the thing you wanted. A toddler can operate an iPad — no manual, no mouse, no syntax. The interface had become so intuitive it was nearly invisible.

Each step has erased a layer of abstraction between what you want and getting it done. So here is my prediction for the next step, and I don't think it's a bold one: the interface disappears entirely, and an AI sits in its place.

"What would you like to do today?"

Imagine you sit down at your machine and instead of a desktop full of icons, you're greeted with a question: What would you like to do today?

You answer in plain language. I need to put together the quarterly report. And then everything assembles itself around that intent. The AI pulls up last quarter's figures, opens the relevant notes, drafts an email to the colleague you always loop in, spins up a browser tab already logged into the analytics dashboard using credentials from your keychain. You do the actual thinking part — the part that's irreducibly yours — and when you're done, the AI files everything away, updates the records, and tidies up after you.

This is not science fiction. I already live a slice of it. I use Claude Code, Anthropic's command-line interface for Claude, as my note-taking system. I tell it "take a note about X." It listens to my half-formed, rambling thoughts, converts them into a coherent, human-readable document, and saves it in an organised structure. Crucially, it maintains an index. It then syncs everything to GitHub, so my notes are live on my phone through the GitHub app moments later.

The magic isn't the saving. It's the retrieval. I can later ask, "what did I conclude about that database migration?" and it searches the index, finds the relevant notes, and extracts exactly the contextual answer I need.

Think about what this fixes. We all have the same graveyard: a notes app stuffed with hundreds of entries where we only ever touch the most recent three. Everything else just sits there — unsorted, unsearched, effectively lost. Static text we'll never read again. My setup turns that dead archive into a living, queryable structure of context that stays organised on its own. I have stopped doing computer admin almost entirely. It has made my life genuinely, measurably easier.

Now picture that, but for your entire machine. Every file, every email, every app, every login — mediated by something that understands what you're trying to accomplish. It sounds wonderful.

And it is.

Or so it seems.

The part nobody puts on the marketing page

Here's the thing I keep circling back to, the reason I actually sat down to write this.

This whole future runs on tokens.

When you talk to an AI model, your words don't go in as words. They're broken into tokens — chunks of text the model can process. The model thinks in tokens, and it answers in tokens. And tokens are the unit you pay for. Every request in, every word of reasoning, every word out: metered.

This is already how the economics work. Claude's usage operates on a session window that resets every five hours, and how fast you burn through that allowance depends on the length and complexity of your conversations, the features you use, and which model you're talking to. The more capable "thinking" models spend more tokens reasoning before they answer — better output, higher cost. And the limits are real: heavy users routinely hit the wall mid-task, because the meter tracks tokens, not the number of messages you send. Run dry and you're locked out until the window resets, or you start paying consumption-based rates on top of your subscription.

Right now, that's fine. The AI is one app among many. When I hit a limit, I shrug and go use a different tool, or I just wait.

But follow my prediction to its conclusion. If the AI becomes the interface — if it's the thing standing between you and your computer, your tablet, your phone, your car, your fridge — then there is no "different tool" to switch to. The AI isn't an app you open. It's the door you walk through to reach everything else.

Which means every single interaction with "your" device now costs tokens. Want to open a file? Tokens. Reply to a message? Tokens. Ask where you put that document? Tokens. The trivial and the profound, all metered the same way. Using your own computer becomes a transaction with a third party, every time, regardless of the task.

Pay to play

We have been quietly trained for this. Look at what we used to own and now merely rent:

Music became Spotify. Films and television became Netflix and a dozen others. Books became Kindle Unlimited and Audible. Your files became cloud storage subscriptions. Even your fitness — the act of doing a workout in your living room — became a monthly fee. Software you bought once, in a box, became Software as a Service that bills you forever and stops working the day you stop paying.

The pattern is always the same: a thing you owned outright is converted into a stream of payments you can never finish making. Ownership becomes access, and access has a meter on it.

The AI-first device is the final and most complete expression of this. It's not a subscription to content anymore. It's not even a subscription to software. It's a subscription to the act of using your own machine. The toll booth moves from the things you do on the computer to the simple fact of touching the computer at all.

I want to be clear: I am not a doomer about this. The technology is real, it's here, and I use it every day precisely because it's so good. The note-taking setup I described has genuinely changed how I work. I'm not arguing we should reject it.

I'm arguing we should walk into it with our eyes open. The most seductive technologies are the ones that make life so frictionless you stop noticing what you traded for it. Every previous interface revolution gave us something and asked for nothing but the price of the hardware. This one is different. It makes the friction vanish — and quietly installs a turnstile where the friction used to be.

So when the day comes that your computer greets you with "What would you like to do today?", it's worth remembering that the question has a second, unspoken half:

…and how would you like to pay for it?