Thoughts on AI Slop and Writing

We live today in the age of slop. AI generates an endless amount of it.

In code - this looks like an explosion of lines.
The proliferation of one line functions wrapping other one line functions.
Five fallbacks for normalizing input.
50 lines of unit tests for every line of code.

For the most part, this has not bothered me.
Because the code is not the point.
Code is the means to an end.
The end is creating software that solves problems for real people.
And if AI slop helps accelerate this process, then so be it.

AI slop in writing bothers me greatly.
Specifically, this is humans trying to pass AI slop off as their own.
It’s up there with Siri interrupting a conversation when she isn’t wanted.

I’ve been reflecting on why this is the case - why the same slop in a different context invokes such a different emotional response.

To answer, I go back to my previous claim that code is not the point in software.
Code is the implementation.
The end result is software that helps people.

Writing is different.
Many times, writing is the end.
It’s the distillation of a thought.
A point of view.
A statement about what we believe.

When people substitute their views with slop - they are making a statement about themselves. And their readers.

Engaging with writing requires that we spend what is becoming our most limited resource - our attention. To spend this precious resource and to discover that what we are reading is something someone else never bothered writing in the first place feels like a gross violation of trust.

It could be argued that writing is also a means to an end.
To change a sentiment.
To transfer knowledge.
To make a point.

I’m sympathetic to this view and I think it is possible to be thoughtful in using AI to create good writing. Just like there is a difference between vibe coding and agentic engineering, there can be a difference between a thoughtless prompt and a well designed system for creating good content.

When I write code, I don’t just tell codex to “make it better”.
I have a carefully curated system of skills, context, and human/agent review loops to guide code slop into good software. The engineers that I respect all have similar systems.

Maybe a sufficiently well tuned system for writing will arise that can shake off the bad Siri vibes from current examples. Even so, there’s still going to be a lingering issue with authorship.

With code, we know that the code is written by an agent.
There’s a human in the loop but that human is not claiming agent generated code as their own.

With writing, this is not the case.
Social media is full of people trying to pass AI generated slop as their own.
As slop gets more sophisticated, this will get harder to detect.

Clear writing reflects clear thinking - but if the writing is not your own, what does it reflect exactly? There is a reason that plagiarism is one of the strongest taboos in academic circles.

It’s hard to see how this will ultimately play out. That said - it’s clear we need some sort of convention to mark content that is solely agent generated vs that which is human authored. How we should treat content that is a mix of both remains an unfolding social experiment.