Why AI writing is still a mess: we're dividing a task we never broke down
Part 1: The 4 components of modern writing
It’s official. Content writing has been commoditized.
If you’re a writer, or a content writing company, move on up the stack to strategy, consulting or other advisory work, because there’s no value in writing anymore.
Thank you for your services. Goodbye now.
That’s what you’d hear if you tuned into the thought leaders of the content industry today.
The argument is that, thanks to AI, writing is now “solved”.
Here's the logic. Say you need to get some writing done, and you’re not a great writer. In the past, you probably would have paid a writer to do it, because he or she would do a better job than you could. And if you wanted it even better, you'd pay an even stronger writer more.
In other words, better writing cost more, and people paid for it.
But now? You just use AI. It's instant, basically free, and the result is as good as what you'd get from the writer anyway. So why pay one? The skill that used to be worth paying a premium for isn’t better anymore.
That’s what they mean when they say commoditized: the expensive version stopped beating the cheap one, so there's no reason to provide that expensive version anymore.
True?
Maybe partly (I’ll get into that in another article), but for the most part, no. It is not. At least, not in my opinion.
If you’re working in and around content day-to-day, you know this yourself. You see the writing of today and it’s often just average.
And I’m not just talking about the pure AI stuff. Even most of the hybrid writing I see, where humans work iteratively alongside AI, still isn’t that great.
It’s often bland and not that interesting to readers. That’s the obvious part. But the process is usually a mess, too. Users end up in long back-and-forth sessions with AI, rewriting, re-prompting, and editing draft after draft, trying to shape something generic into something that actually feels like it’s worth publishing.
That or they just give up and ship what they have. Either way, it's a lot of work for a result that's just okay.
That doesn’t sound like a solved problem to me. Rather, it sounds like one in its infancy.
And why wouldn’t it be? We’re only a few years into AI even existing, so it makes sense that the way we write with it is immature.
Now, don’t get me wrong, I’m not an AI hater. I’m all for using AI to help with writing. I really am. I think there’s a beautiful future where humans and AI work together to create good content in an efficient manner.
We’re just not there yet. The modern writing process (one where we use AI) can be improved greatly from where it is now.
The big question, of course, is how? How do we go from this informal, messy hybrid process that yields inconsistent results to something organized and predictable that produces content people actually want to read?
The answer, I believe, lies in how we divide the work between human and AI. I wrote about this idea in my last article, AI doesn’t suck at writing, you’re just giving it the wrong job.
That piece set a foundation for improvement, but led to a new question. A more specific one. One that might explain at its roots why our current AI-writing efforts are falling short.
If modern writing is a hybrid task, then the labor needs to be allocated somehow. Obvious. Okay. But to allocate a task well, you first need to break it down into its parts. And as far as I can tell, we haven't really done that with hybrid writing.
So my question is, can writing actually be broken down at a more granular level? Because if it can be, maybe that's our first step. Maybe that's how we move from messy to organized and from generic to compelling. Maybe that's how we ensure AI is doing the right parts of the job, and actually helping writers instead of just getting in the way.
You can’t divide what you haven’t broken down
Before we go any further, it’s important to define what we’re dealing with here.
We're looking at the process of human and AI working together to write content. Some people call it hybrid writing. I like to call it modern writing. Whatever you call it, it's a collaborative task being worked on by more than one contributor.
That last part is key, because it makes modern writing fundamentally different from writing as we've known it. For hundreds of years, writing was a monolithic task done by a single person. One mind, one undivided job, from blank page to draft.
Yes, there's always been collaboration around the writer in the form of editors, fact checkers, or subject matter experts, but those roles supported the writer. They couldn't write with, or for, the writer. Nothing about those arrangements was as fluid, real-time, or powerful as what we're dealing with now.
So when I refer to "writing" here, I mean the act of turning raw ideas into readable words on a page, not the supporting process around it.
And when you look at writing that way, you realize it’s undergoing a monumental change. We’re moving from something that was a single-person task for hundreds of years to a collaborative one, almost overnight.
Simply put, this changes the game.
The problem is, we’re still playing it the same way. We’re still acting like it’s a one-contributor job.
We need to update our strategy to match the new game in town.
The good news is, there's an obvious place to start.
Think of it this way. In any collaborative task, after we know who’s going to be involved, we always ask one important question: how should the work be divided between them?
Because when the division of labor is clear and deliberate, the work gets better, faster, and easier. When it isn't, the whole process struggles, no matter how capable each contributor is.
But with hybrid writing, we’re treating the whole thing as one blended effort. And that’s just bad practice when doing any kind of collaboration.
Look at a restaurant kitchen. You don't have every cook making every dish from start to finish. Someone preps, someone grills, someone plates. The work gets broken down and matched to whoever is best for each piece. It's the same in construction, in software, in medicine, in just about every other field where multiple people work on something complex together.
Now imagine the opposite. Every cook doing every job, no roles, no breakdown, everyone bumping into each other trying to make the same dishes. It would be a disaster. That's basically what modern writing looks like right now.
If we want writing to be better, we need to figure out how to divide the work between human and AI in the smartest possible way.
But here's the funny thing about all this.
You can't actually divide something without breaking it into parts. Isn’t that kind of a requirement to the whole process?
You decompose, then you allocate. That’s the order of operations.
Where we went wrong? We skipped the breakdown part entirely.
We jumped straight into hybrid workflows, putting humans and AI side by side and splitting the work between them on the fly in arbitrary ways, without ever stopping to ask what writing is actually made of.
We're still treating it as a single monolithic task while trying to spread that single task across two contributors, and it leads to exactly the kind of messy, inefficient process you'd expect. Just like those cooks stepping on each other's toes.
We're trying to divide the work of a task we never broke down in the first place.
In my last article, I proposed a high-level framework to help us divide the labor of writing. I argued that humans should own the "what" and AI should help with the "how". I still think that’s the right starting point. It works as a compass, pointing us in the right general direction.
But a compass on its own isn't enough. We need a map.
If the future of writing really does depend on how well we distribute the work between human and AI, we need to break writing into its fundamental parts and understand how each one works. Only then can we start matching those parts to the strengths of human and AI in a deliberate way.
We tried to run before we walked. So let’s take a step back and do it right this time.
We've been breaking writing down for a long time, just not in the right places
Okay, we know we need to break writing down before we can allocate it properly between human and AI. But where do we start?
Lucky for us, our friends in the editing world have been breaking writing down for a long time. Their whole system is built on the idea that reviewing different parts of writing requires different skills.
Maybe we can pick up a few clues from their approach.
Here's how editing is typically divided:
- Developmental Editing (also known as Structural or Substantive Editing): The big-picture pass. Focused on ideas, arguments, organization, and overall substance. Does the piece say something meaningful? Is it built right?
- Line Editing (also known as Stylistic Editing): The sentence-level pass. Focused on clarity, flow, tone, and wording. Does each sentence land the way it should?
- Copy Editing: The correctness pass. Focused on grammar, punctuation, spelling, and consistency.
- Proofreading: The final clean-up. A last look for typos, formatting issues, and small errors before the work goes out the door.
It's a great breakdown, and it aligns closely with how I think we should break down writing for hybrid creation. You can already start to see how each layer could map to the human what / AI how framework.
The catch is that this model was built for reviewing and improving writing after it's already been done. We're trying to break things down for a different purpose. We're trying to allocate the work of writing while it’s being done.
Remember, we're dealing with something completely new here. For the first time, another entity can participate directly in the act of writing itself. Not just reviewing it afterward, but actively helping generate and shape the writing in real time.
That changes what we need from a decomposition model.
The biggest issue I see with this model is that it bundles two very different things together at the top. Developmental editing treats ideas and the organization of those ideas as one job. But they're not. You can have great ideas in a terrible structure. You can have a perfectly organized piece built on weak ideas. Treating them as one layer hides a distinction that matters when you're trying to assign the work of writing in real time.
Another model worth looking at is classical rhetoric, which goes all the way back to thinkers like Aristotle, Cicero, and Quintilian. Their model was built mainly for speaking, not writing, but it still includes some useful components.
The five canons of classical rhetoric are:
- Invention: Coming up with the ideas and arguments
- Arrangement: Organizing those ideas
- Style: The wording and language used to deliver them
- Memory: Committing the speech to memory
- Delivery: The voice, gestures, and presence used in actually performing it
What's interesting is that, like the human what / AI how framework, this model also separates ideas (invention) from the way they're expressed (arrangement and style). So even thousands of years ago, people recognized that what you say and how you say it are different things. Cool.
But it also includes parts that don't really apply to writing, like memory and delivery, which are specific to the act of speaking.
So neither model is a perfect fit. The editing model is close but lumps too much together at the top. Classical rhetoric has good bones but includes layers that don't translate to writing.
We need something more granular and complete that takes into account how writing actually works today.
The four components of modern writing
What I'm proposing is a four-component model that's grounded in both of those traditional versions but updated to align with the current writing landscape:
- Ideas: The substance of the piece. The claims, arguments, examples, and meaning the writer wants to convey.
- Structure: The organization. The order ideas appear in, how the piece flows, where the emphasis sits, and how the parts connect.
- Expression: The act of turning raw ideas into language on the page. The wording, the phrasing, the tone.
- Mechanics: The technical layer. Grammar, punctuation, formatting, and consistency.
You'll notice these four components align closely with parts of the other two models, with a few important differences.
The biggest one is that we've kept ideas on their own.
What I love is that classical rhetoric actually had this right. Way back then, those great thinkers decided that invention and arrangement were separate canons. Coming up with ideas and organizing them were treated as two different kinds of work. They got it.
The editing world bundles them together under developmental editing, and to be fair, that probably makes sense in their context. Editors aren't usually the ones generating the core ideas of a piece. The writer brings the substance. The editor's job is mostly about shaping how that substance comes through, which means most of their work in this layer is on the organization side. If the ideas themselves are weak, that's typically a different conversation. So treating ideas and organization as one job works in that context.
But we're looking at the act of writing itself, where ideas are a huge part of the process. And when you throw AI into the mix, the distinction between ideas and structure becomes even more meaningful.
Why? Because structure can now be worked on independently from ideas, and vice versa. AI can organize, sequence, outline, and restructure human-provided substance. It can also suggest ideas. This doesn't mean AI should handle either layer on its own. It just means both can now be handled separately by either AI or human. And because that's possible, the two layers should be broken out. Otherwise, you might miss who's doing what, and the option to allocate them disappears before you ever get to consider it.
That's why ideas and structure get their own slots in this model. Not because the distinction is new, but because the two can now be allocated separately.
The other difference worth flagging is expression. It maps closely to the line/stylistic phase in editing and to the style canon in rhetoric, but I think "expression" is a more accurate term for what's being done here.
Expression is the foundational act of turning raw ideas into language. Style implies a further step on top of that, where you shape the language into a specific flavor for a desired impact. Style still matters. It's just a layer on top of expression, not the same thing.
Finally, mechanics is simply the concept of correctness, of following the rules of language. It maps cleanly to the copy editing phase and focuses on technical things like grammar, punctuation and consistency to make the reading process smooth and frictionless.
When you put it all together, it’s a simple breakdown:
- You start with raw ideas
- You figure out how to organize them (structure)
- You turn them into language (expression)
- And you make sure the language is technically correct (mechanics)
This is the natural process of writing. It's always been there. It just hasn't been explicitly broken out like this before, because for most of history it all happened fluidly inside one person's head.
And if you’ve been following along, you might have noticed how cleanly these four components line up with the human what / AI how framework we started with. You have ideas, which make up the “what”, and then you have structure, expression, and mechanics, which make up the “how”.
This puts us exactly where we want to be. We now have the map we needed and we know it aligns with our original compass.
Using the four components to make writing better
Now that we have writing broken down cleanly into its parts, we can safely move to the next step of figuring out how to best distribute the work.
Is it a perfect model? I don’t know yet. But it's a starting point. And that’s what we’ve been missing.
Because although most people today believe the future of writing is humans and AI working together, that belief alone isn’t actionable. It doesn't tell us what to do differently tomorrow.
These four components might. They provide a vehicle to assign writing labor in an organized way. The enabling layer we skipped the first time around.
From here, the big question becomes obvious: who should do what? Between human and AI, is there an ideal fit for each component?
Out of ideas, structure, expression, and mechanics, does either human or AI handle some parts better than others? The answer is almost certainly yes.
That's what I’ll be exploring in Part 2 of this article. We’ll look at the natural strengths and weaknesses of humans and AI and see how they match up against each of the four components.
And we'll touch on an even bigger question that shows up once you start looking at writing this way. If writing really can be broken into components, and each one can be worked on independently, then why are we stopping at two contributors? Instead of a single human writing with AI, can we do it with multiple humans in parallel?
The promise of hybrid writing has always been that it could be faster, smarter, and better than what came before. We're not there yet. But with it now broken down, like any collaborative task should be, we're finally in a position to look at the problem critically and move forward.