JOURNAL

Notes from the Dark

What I am learning as The Failing Light takes shape.

Browse entries

How AI Supports My Process

Did AI do most of the work building this website?

Yes.

Did AI write The Failing Light?

No.

Good, you're still reading.

There are plenty of people who are completely against the very idea of AI. They're probably long gone by now. And that's alright. I'm not trying to run anyone off, but I'm also not trying to change any minds.

So I'm just going to lay this all out.

AI is deeply integrated into how I develop The Failing Light.

I use it frequently.

I use it to challenge my thinking, test structure, track continuity, research craft and psychology, compare alternatives, find contradictions, and diagnose problems in the work.

But more than anything else, I use it to help me organize all of the worldbuilding and supporting material behind the novel.

I have a very architectural approach to writing. I do a lot of foundational work and planning before I'm ready to start drafting prose. And I'm constantly throwing new ideas into the mix.

I watched a series of Brandon Sanderson's writing lectures online that I really enjoyed. One idea from it that stuck with me was essentially this: use your cool ideas. Don't save all the good stuff for some hypothetical future book. Put it into the thing you're writing now.

I took that to heart.

Sometimes a new idea slots perfectly into the architecture I've already built.

Sometimes it's like chucking a grenade into my living room.

Helping me keep the core themes and concepts locked in while I explore those forks in the road, then keeping the consequences organized and documented afterward, is probably the biggest way I've leveraged AI.

But there have to be boundaries.

AI does not draft my chapters.

Frankly, I think AI is terrible at writing prose.

Seriously, try it sometime. It's absolute trash.

I also don't hand it authority over my scenes, outlines, or roadmaps. It can help me test them. It can challenge them. It can urge me to try other paths, and it can point out what I might break if I change something.

But it doesn't decide what's canon.

It doesn't decide what my characters think or feel.

It doesn't decide what the story means.

It doesn't invent new lore or mechanics just because they would conveniently solve some problem I'm stuck on.

It can challenge my decisions.

It cannot make them for me.

The manuscript is mine. The creative decisions are mine. The final judgment about what becomes true is mine.

What AI has become is part of the infrastructure around the book.

It helps me focus, test, revise, organize, remember, compare, and occasionally realize that I have spent three days protecting an idea that should have been cut on the first.

This website is another good example.

AI did most of the work required to build it.

I decided what I wanted it to be. I decided what it should say. I decided what belonged here. But I had very little interest in spending weeks teaching myself web development just to prove that I could personally type every line of HTML.

So I used the tool.

But a novel is not a website.

There are parts of making The Failing Light that I specifically do not want to outsource, because struggling with those parts is the work.

I need to decide what secrets a character is keeping.

What scene needs to happen in Chapter 1 so that something in Chapter 30 lands the way I want it to.

When a small detail that reads one way on the first pass should mean something completely different on the second.

I need to recognize when I'm protecting something in the manuscript because it serves the thematic core of the book, and when I'm protecting it because I just really (really) like what I wrote.

AI can help expose those questions.

It can never be trusted to answer them for me.

If I took the best chapter I had ever read from the best book ever written, uploaded it to AI, and asked how to make that chapter better, it would give me a list of things.

An entire list of garbage.

And it would have no idea the list was garbage.

You could challenge that list and tell the AI you think it's garbage. And it would respond:

You're right to challenge me. After reconsidering, I have to agree that the entire list I gave you was, in fact, total garbage.

You can't trust it.

You have to be the one who decides whether its answer is useful, irrelevant, or garbage.

I'm not interested in minimizing AI's role in this project.

I am equally uninterested in confusing that role with authorship.