Do AI Writing Tools Make Typing Obsolete?

Producing text is the job everyone means when they say typing: turning a sentence you have in mind into characters on screen. This is the part generated text genuinely reduces, and for some kinds of writing it reduces it a great deal.

Specifying is the job of describing what you want — a prompt, a brief, a comment explaining an intention. This is not reduced; it is created. It did not exist as a separate keyboard activity twenty years ago and it is now a substantial share of some people's typing.

Revising is reading output and changing it: adjusting a phrase, deleting a paragraph, correcting a fact, moving a section. Generated text increases this, because it produces a great deal of material that needs judgement applied to it.

So of the three, one shrinks and two grow. Whether total keyboard time falls depends on the ratio for your particular work, and for a lot of people it does not fall at all — it changes shape.

The two growing jobs share a property that the shrinking one does not: they are short, and they are unforgiving.

A thousand-word document tolerates a typo. It will be caught, or it will not matter, and the meaning survives either way. A forty-word instruction does not tolerate one nearly as well, because a wrong word can change what gets produced — and you may not notice from the output that it was your input that was wrong.

Revision is the same. Editing is a sequence of small precise operations on existing text, each of which has to land exactly where intended, and an error there is a change you did not mean to make in a document that already looked finished.

So the load moves from volume to precision. That is exactly the direction in which accuracy work becomes more valuable rather than less, for the reasons totalled up in accuracy versus speed — and it is the opposite of what "typing matters less" implies.

If more of your keyboard time is revision, then more of it is cursor movement, selection and replacement rather than continuous typing.

That has an awkward implication: the skill that matters most is the one nobody measures. Moving by word, selecting to the end of a paragraph, deleting a phrase, jumping between two positions — none of that appears in a words-per-minute figure, and all of it is what editing consists of.

A person who can produce ninety words a minute of new prose but navigates a document with a mouse is badly equipped for a workflow that is mostly revision. Someone slower who never leaves the keyboard is better equipped, and the gap widens as the proportion of revision rises.

This is the argument in shortcuts are the other half of speed, and generated text strengthens it considerably. If the production half is shrinking, the navigation half is a larger share of what is left.

It is worth listing the keyboard work that generation does not touch at all, because it is more than people assume.

Anything with a hard correctness requirement and no natural language in it: passwords, reference numbers, account details, command lines, spreadsheet cells, database fields. There is nothing for a model to generate and no forgiveness for an error, which is the profile described in typing for data entry.

Anything conversational and immediate: a support chat, a live message thread, a meeting note taken while someone is talking. Speed and immediacy are the point, and a generation step in the loop costs more time than it saves.

Anything private or offline. And anything where the writing is the thinking — where you find out what you meant by writing it — which is a substantial fraction of real work and is not a task that delegates.

Across all of those, the keyboard is exactly what it was, and the person who cannot use it well is exactly as disadvantaged as before.

This is not the first technology forecast to be made about typing. Dictation has been reliably predicted to replace it for decades, has become genuinely good, and has not replaced it.

The reason is worth understanding because it applies here too. Dictation is faster than typing at producing words and slower at producing finished text, because it handles punctuation badly, handles formatting worse, cannot be used in shared spaces or on anything confidential, and demands a level of composed-before-you-speak fluency most people do not have. The full comparison is in voice-to-text versus typing.

The general lesson is that input methods do not replace each other; they partition the space by task. Dictation took the tasks it is good at — long-form first drafts, hands-busy situations — and left the rest.

Generated text will do the same. It is taking the tasks where a competent first draft from a description is what you need, and it will leave the rest, which is a large remainder.

If I were advising someone allocating limited practice time now rather than five years ago, three things would change.

Accuracy over speed, more strongly than before. Short precise input is the growth area, and an error in forty words costs more than an error in a thousand. The zero-error drill and the no-backspace test are more relevant than another speed session.

Navigation over production. If revision is a growing share of keyboard time, editing efficiency is a growing share of keyboard value, and it is cheap to acquire relative to speed.

Symbols and numbers, unchanged in importance and possibly increased. They are the material generation cannot help with, they are what precise fields are made of, and almost nobody has practised them.

What I would not advise is abandoning typing practice on the grounds that the machine will do it. The machine does one of the three jobs, and it makes the other two matter more.