After the Keyboard

Edwina McKennon’s video, GPT-6 Astra Is the End of Computers, begins with a piece of footage from MIT in 1979. A man points at a screen and tells the computer what to do. The system was called Put That There, and the idea behind it was simple: we shouldn’t have to learn the computer’s language; the computer should learn ours.

Nearly half a century later, we’re still working our way towards that idea.

Edwina reads OpenAI’s use of the footage as a signal. First we operated the machine. Then we learned to talk to it. Now we’re approaching the point where we decide what needs doing and the machine handles the machinery.

I think she’s right about the direction of travel.

I already feel it in my own practice. I dictate many of my thoughts through Wispr. I talk with AI agents that search, retrieve, organise, draft, and carry out work across different systems. I’m spending less time clicking through menus and more time trying to describe what I want, why I want it, and what a good result would look like.

The interface is beginning to recede.

That doesn’t mean the keyboard will disappear tomorrow. People have been announcing the death of paper since the arrival of the word processor, and my desk still accumulates notebooks as if trees are personally sending me correspondence. Technologies rarely vanish on schedule. They linger, find niches, become aesthetic choices, or survive because they offer something the replacement doesn’t.

Typing, for example, isn’t merely a slow way of entering information. It can be a way of thinking. The pace of the fingers, the pause before a sentence, the visible trail of revision—these are part of the cognitive process. Speaking produces different thoughts from writing by hand, and writing by hand produces different thoughts from typing. Removing friction can create freedom, but friction sometimes gives thought the surface it needs to take shape.

So I’m less interested in whether the keyboard is literally doomed than in the deeper movement Edwina identifies: operational competence is losing some of its scarcity value.

For years, knowing how to make the software behave was a marketable skill. You knew which menu contained the command, how to construct the spreadsheet, how to format the slide deck, how to translate an intention into the particular grammar demanded by a machine.

Now the machine is becoming responsible for more of that translation.

Prompt engineering may turn out to have been a transitional craft: a temporary dialect we learned while computers were getting better at understanding ordinary language. But the disappearance of elaborate prompting won’t make articulation irrelevant. Quite the opposite. We may no longer need incantations full of parameters and special syntax, but we’ll still need to know what we mean.

That is a much older and more difficult skill.

Edwina names judgment and taste as the capacities gaining value: knowing what needs to be made, why it matters, and whether the result is any good. I’d add attention, context, and responsibility.

Attention notices what the machine missed.

Context supplies what the machine can’t infer from the public internet.

Taste distinguishes something technically adequate from something alive.

Responsibility answers for what enters the world.

The machine can produce a polished document without knowing whether the document should exist. It can satisfy the written brief while violating the deeper intention. It can fill gaps so fluently that we don’t notice a decision has been made on our behalf.

That last possibility is the part of the video that stays with me.

Edwina points out that when we leave something unspecified, the system guesses. Increasingly, it won’t interrupt us over every ambiguity. It will decide which gaps are routine enough to fill for itself.

This is what makes AI feel useful. It is also where its quiet power gathers.

Every inferred preference, completed field, chosen phrase, or unasked question nudges the work in a particular direction. Most of these choices will be harmless. Some will save us hours. But convenience can conceal authorship. A decision doesn’t stop being a decision simply because we didn’t notice it happening.

Edwina offers four questions before handing work to AI:

1. What should the finished result look like?
2. What does the machine need that it can’t safely guess?
3. What requires my approval before it goes out?
4. How will I know the result is right?

That’s a useful test. I might add a fifth:

What part of this work do I need to remain in contact with?

Not every difficulty is waste. Not every task should be delegated simply because it can be. If the work helps me notice, learn, remember, practise judgment, or discover what I think, handing it over may cost more than it saves.

Writing is the obvious example for me. An AI can help me search the archive, test a structure, expose a weak assumption, or carry a fragment into dialogue. But if I outsource the encounter itself, I may receive a competent piece of prose while losing the thinking the piece was meant to make possible.

This is the distinction I want to preserve as computers disappear into the background: the difference between avoiding machinery and abandoning participation.

There is a liberating possibility here. If the screen no longer demands that I sit before it, perhaps I can return to the garden, the path, the train, the café, and the ordinary traffic of life while the machinery takes care of its own menus. The computer becomes less of a destination and more of a companion moving quietly at the edge of the scene.

But there is another possibility. Work follows us everywhere because there is no longer even a desk we can leave.

The disappearance of the computer doesn’t tell us which future we’re entering. That remains a question of design, culture, power, and choice.

The interface may recede. Our responsibility doesn’t.

And perhaps that is the real work waiting after the keyboard: learning not simply how to command more powerful machines, but how to remain awake while they act.