AI Is Not Driving My Car. It Is Power Steering.

There is a particular way people sometimes talk about AI and writing that has never quite matched my own experience.

The comparison is usually implicit, but it is there: using AI to write is like getting into a self-driving car.

You tell it where you want to go. You sit back. The machine does the work. You arrive at the destination and then claim that you drove there.

I understand why people make that comparison.

More importantly, I think there are people using AI exactly that way.

There are books being produced by people who type a prompt, let a system generate large amounts of text, make minimal changes, and publish the result. The criticism of low-effort “AI slop” did not appear out of nowhere. There is plenty of material out there that deserves that criticism.

I am not particularly interested in defending that process.

But it is not my process.

For me, writing with AI is not a self-driving car.

It is power steering.

And I think the distinction matters.

The Wheel Has Been Getting Easier to Turn for a Long Time

Imagine writing a novel a century ago.

Pen and paper. Longhand drafts. Physical pages. If you wanted to move a paragraph, you rewrote it. If you wanted to change a chapter substantially, you might rewrite pages and pages simply to produce a clean copy.

Then came typewriters.

They did not write the book for you, but they changed the labour involved in producing it.

Then word processors.

Suddenly moving a paragraph did not mean rewriting three pages. Searching for a word did not mean physically scanning an entire manuscript. Whole sections could be moved, copied, deleted and restored in seconds.

Then spellcheckers. Grammar tools. Digital dictionaries. Search engines. Electronic notes. Cloud storage. Version histories.

Each step removed another layer of friction.

And those steps were not equal.

The word processor was not simply a slightly better typewriter. Spellcheck did something a sheet of paper could not do. Modern grammar tools do substantially more than the first red squiggly line under a misspelled word ever could.

The progression has accelerated.

AI is a much bigger leap again.

I do not dispute that.

What I dispute is the idea that because the assistance has become dramatically more powerful, authorship must automatically have transferred to the machine.

Power steering is also a significant mechanical intervention.

In an old car without it, turning the wheel at low speed can require real physical effort. The driver fights the weight of the vehicle through the steering mechanism.

Power steering adds force.

A small movement from the driver can now turn something much heavier.

The technology is absolutely doing work.

But it is not deciding where the car goes.

The driver still chooses the road.

The driver still decides when to turn.

The driver still decides when to stop.

And if the driver puts the car into a ditch, blaming the power steering would be rather strange.

That is much closer to what AI means in my writing process.

I Still Have My Hands on the Wheel

I never ask AI to write a book and then move on.

I do not ask it to write a chapter and simply accept the chapter.

I do not even ask it to rewrite a paragraph and then assume the rewrite is better because a machine produced it.

There are certainly times when I will give AI a paragraph I have written and ask it to make the meaning clearer.

Then I read the result.

I change it.

I feed the changed version back through.

I see what happens.

I change it again.

Sometimes that process happens several times before I am satisfied.

The final authority is still me.

Not because the AI contributed nothing. That would be dishonest.

It contributed assistance.

What it did not contribute was the final decision.

That distinction runs through almost everything I use it for.

I know, for example, that I have a tendency when writing quickly to become attached to particular descriptive words. I can use the same kind of phrase repeatedly without noticing because each occurrence made perfect sense when I wrote it.

AI lets me interrogate that problem quickly.

How often have I used this word?

Where?

What alternatives might fit these particular sentences?

I can then look at those suggestions and decide which repetitions are intentional, which are invisible problems, which alternatives sound like me, and which absolutely do not.

The machine does not decide what my prose should sound like.

It makes it much easier for me to see the prose that is already there.

Sometimes I Know What I Mean Too Well

One of my recurring problems as a writer is that I know the story.

That sounds useful.

Sometimes it is not.

A scene can exist so clearly in my head that I forget the reader cannot see it.

I know why a character crossed the room.

I know what happened three chapters earlier.

I know that two sentences refer to separate ideas.

I know a particular piece of worldbuilding is true.

And because I know it, I can read a passage and unconsciously supply information that is not actually on the page.

That is where AI has become enormously useful to me.

I can give it a passage along with the earlier material and ask something very simple:

What do you not understand?

What assumptions are you having to make?

Is there anything happening here that the text has not actually established?

That is not so different from a beta reader saying, “Wait. How did she know that?”

The difference is that I can get the question in seconds rather than days or weeks.

The answer is still mine.

If the AI does not understand something, I have to decide whether that confusion matters.

Maybe I deliberately want ambiguity.

Maybe the answer is coming later.

Maybe the AI missed something obvious.

Or maybe I suddenly realise that the reader is being asked to know something I never actually told them.

That last one happens more often than writers would probably like to admit.

AI can see the page.

It cannot see inside my head.

That makes the gap useful.

It Can Also Be Wrong

This matters just as much.

While writing A Funeral for the Living, AI assistance repeatedly pushed toward faster pacing.

More action.

Quicker escalation.

Less time sitting with characters.

More obvious movement.

And I repeatedly said no.

The story was intentionally slow-burning.

I knew where it was going.

The AI knew what I had shown it and could make reasonable observations based on that material. But it did not possess my complete understanding of the story.

It could see the road immediately ahead.

I knew where the road was going.

Had I treated every suggestion as an instruction, I would have damaged the books.

So I rejected those suggestions.

Again and again.

That is authorship.

Not the absence of outside input.

The exercise of judgment over that input.

Writers have always received suggestions. Editors suggest things. Beta readers suggest things. Friends suggest things. Other writers suggest things.

Sometimes they are right.

Sometimes they are spectacularly wrong for the story you are trying to tell.

The writer still has to know the difference.

AI has not relieved me of that responsibility.

If anything, its speed makes that responsibility more important.

Faster Does Not Mean Less Work

One criticism of AI-assisted writing that I find particularly interesting is the assumption that because something becomes faster, the writer must therefore be doing less.

That has not been my experience at all.

There are days when I spend twelve or fourteen hours writing, reading, revising, checking, restructuring and thinking.

AI has not made those days disappear.

It has changed how much I can accomplish during them.

Without that assistance, a book might take me six months. A year. Longer.

With it, I may finish significantly faster.

But the time saved does not become empty time.

It becomes the next paragraph.

The next chapter.

The next editing pass.

The next book.

That is what I mean when I call AI a force multiplier.

It does not remove my effort.

It allows more of that effort to reach the page.

The purpose of the tool is not to reduce my contribution.

It is to concentrate my contribution where it matters most.

Speed Can Improve Continuity Too

There is another benefit that surprised me.

Writing faster can actually help me maintain continuity.

If Chapter Two was written six months ago, I may need to spend real time reconstructing my own mental state from when I wrote it.

What exactly was this character feeling?

How had their relationship changed at that point?

What promises had I made to the reader?

What did this particular conversation mean emotionally?

Of course I can reread it.

I can consult notes.

I can rebuild that state.

But there is a difference between remembering something intellectually and still carrying it.

When Chapter Two was last week rather than six months ago, I often still carry it.

The characters remain closer.

Their emotional state is easier to feel.

The rhythm of their voices is still in my ear.

The unresolved tension has not gone cold.

That means the productivity gain is not merely quantitative.

Sometimes faster means I maintain a stronger connection to the work.

Sometimes faster means I remember exactly who these people were last week.

My Own Language Is Complicated Enough

There are smaller examples too.

I am terrible at spelling.

Spellcheck has been saving me from myself for years.

And yes, I know someone can object that spellcheck is not the same thing as generative AI.

It is not.

That is not my point.

My point is that we have already accepted a long progression of tools that examine what we write and assist us in producing what we intended to write.

Even spellcheck itself has evolved far beyond simply comparing a word to a dictionary.

My own language creates another complication.

I grew up in Louisiana.

I have now lived in the UK for almost half my life.

As a result, my natural vocabulary is an occasionally chaotic mixture of American and British spelling, slang, phrasing and terminology.

That is fine when I am speaking as myself.

It is less fine when a character or setting requires consistency.

AI can help flag when my own linguistic soup has leaked into places where it does not belong.

Again, the goal is not for the system to decide how my characters speak.

The goal is to help me notice when I have accidentally made them speak like me.

This Essay Is an Example

In fact, this essay itself is a good example of what I mean.

I did not ask an AI:

“Tell me what I should believe about AI and writing.”

I started with an idea.

Writing with AI, in my own process, does not feel like using a self-driving car.

It feels like power steering.

I explained why.

The AI reflected back what it thought I meant.

Some of it was right.

Some of it needed refinement.

That response prompted me to explain the idea more clearly.

I added the progression from pen and paper to typewriters, word processors, spellcheck and increasingly powerful assistance.

The AI organised that.

I clarified that I was not defending every use of AI.

It reflected that distinction back.

I supplied examples from my own writing process.

Word repetition.

Copy positioning.

Reader comprehension.

Rejected pacing advice.

Iterative rewriting.

Language consistency.

Time savings.

Continuity.

Each time, I looked at what came back and decided whether it represented what I actually meant.

Sometimes it did.

Sometimes it almost did.

And “almost” is useful, because it makes the difference visible.

That process helped me understand my own argument more clearly.

It did not tell me what to believe.

It helped me make what I already believed legible to someone who cannot see inside my head.

That is very close to how I use AI when writing fiction.

I turn the wheel.

The assistance responds.

Transparency Needs More Than a Warning Label

I do believe authors should be transparent about AI use.

But I do not think a label that simply says:

AI USED

tells anybody anything useful.

There is an enormous difference between:

“I gave an AI a prompt for a novel, accepted most of what it produced, lightly edited it and published it.”

and:

“I used AI to search for repetitive language, interrogate continuity, test reader comprehension, compare passages, assist with editing and iteratively refine prose I was already writing.”

Putting both under the same two-word label erases exactly the distinction we ought to be discussing.

If transparency matters, then the disclosure should contain information.

Tell readers what the tool did.

Tell them what the author retained control over.

Tell them enough to make their own judgment.

And readers, in return, should ideally engage with what that disclosure actually says rather than deciding that the presence of the letters “AI” ends the conversation.

That does not require anyone to approve of my process.

People are allowed to decide that they do not want to read anything touched by generative AI.

That is their choice.

But I would rather that choice be based on what I actually did than on an assumption about what “AI-assisted” must mean.

I Am Not Defending Everything

I want to be particularly clear about this.

I am not asking anyone to approve of every use of AI.

I do not approve of every use of AI myself.

I understand why people are angry about low-effort generated books.

I understand why writers worry about being displaced.

I understand why readers care about authenticity.

I understand why artists are concerned about where training material comes from, how their work is treated, and what happens when automation is used primarily to remove humans from creative work.

Those are real conversations.

What I reject is the idea that the only intellectually coherent response is therefore a complete veto on every use of AI in a creative process.

That position collapses too many fundamentally different behaviours into one category.

For me, the meaningful question is not simply:

Was AI present?

It is:

Who was steering?

Who chose what the story meant?

Who decided who these characters were?

Who rejected the wrong suggestions?

Who decided when a sentence worked?

Who decided when it did not?

Who carried responsibility for the finished work?

If the answer is the author, then I think there is room for a more interesting conversation about tools, assistance, responsibility and authorship than a binary yes-or-no label allows.

I Still Want to Drive

I do not want AI to imagine for me.

I do not want it deciding what my characters believe.

I do not want it deciding what my stories mean.

I do not want it choosing where I am going.

I want it to help me see repetitions sooner.

Catch confusion faster.

Hold more information within reach.

Reduce the mechanical burden of revision.

Give me something external to push against when an idea exists clearly in my head but not yet clearly on the page.

I want to spend less of my finite creative energy wrestling the machinery of writing and more of it making creative decisions.

I still want my hands on the wheel.

The direction is still mine.

The destination is still mine.

The responsibility is still mine.

I simply do not see virtue in making the wheel harder to turn.