AI Writing Sucks

Mel

21 Oct 2026 • 4 min read

Yesterday, I ran into this following blog post:

And the specific snippet that I wholeheartedly agreed on is the last section:

The loss of precision and style affects reading speed. You need to switch ambiguous words for clearer ones while reading, and you have less information about the context to do so, as it is lost due to the lack of style. Personally, I feel that I am at least a third slower reading AI-generated text, even if the length is the same. Beyond this, most models are extremely verbose, which makes total reading time much higher. You do not need many readers for the writer’s time savings to evaporate on the readers’ side. This is even before you take second-order effects into account: more questions, more iterations (if it’s a collaborative text), and less trust between readers and the writer.

At work, I'm a staff/senior level engineer, and oh boy do I do a lot of doc writing and a lot of doc reviewing. There are a few qualities of a well written doc, outside of the technical and engineering aspects, which I value a lot:

The doc must read from top to bottom. This seems obvious as first, but what I specifically mean is that the doc must not require the readers to jump between sections, reread prior sections or jump down to later anchors to "figure out the doc". Every section should be a chronological build-up, so that the author's intent can be fully absorbed even with a top-to-bottom single-pass skim.
The doc must never reference unexplained jargon or ambiguous statements. Non-self-explanatory proses or jargon that requires deep SME knowledge makes docs not "self contained". The moment you expect the reader to have outside knowledge, either you'll spend the entirety of the doc review "explaining" what could have been explained in the doc in the first place, or requires reviewers to spend review time reading other relevant artifacts, wasting precious review time.
The doc must be at that perfect level of technicality. Too low level, nobody understands what you're explaining unless the reviewers know the code by heart. Too high level, the material becomes dumbed down. I personally like to start with a very high level explanation, ie an ELI5 version (or more like ELI20 but let's move on), followed by a slightly lower level explanation where the prior ELI5 version handholds the reader to build out a mental model so that the technical explanation is easier to understand.

Issues with AI writing as a reviewer

I feel like recently a lot of doc reviews have become very unproductive. Sometimes I question if the author have even reviewed their own AI written "draft", because the doc that is under review feels very much like a draft than a publication.

Some giveaways just make me want to stop reading, such Opus finding fancy-pantsy replacement terminology for already established vocabulary. Some that I ran into recently:

"Remedy Ladder" - Retry Strategy
"Trusted Source" - Source of Truth

Like, come on. These are established engineering terminology. When the doc keeps using these new made-up terminology in place of already established ones, I go "WTF is a god damn remedy ladder", and I have to juggle my brain for a whole 5 minutes to realize it's describing a retry strategy.

And here's the thing - not only do I feel like I'm wasting my time, but my impression of YOU as an engineer gets worse and worse.

Engineering docs are how I rate people (even prior to AI)

At the end of the day, the way I rate my engineers (not saying that there is a rating system or anything at my workplace, just my mental tier list of my peers) is based on their engineering capabilities, specifically their thinking/problem-solving process. Have they considered the big picture, or are they tunnel-visioning? Are they prioritizing the proper tenets and requirements, or are they making decisions without being data driven?

Engineering docs are usually the only artifact that documents one's engineering capabilities - especially at an age where code is cheap. If my rating of an engineer used to be maybe 70:30 doc vs CRs, today it's more closer to 90:10.

A well written doc is specifically supposed to be a written version of an engineer's thought process - a document that describes "this doc is how I understood this problem, how I've broken down the problem, the research that I've done, the proposal I have arrived on, and why I believe my solution is the best possible implementation."

It's fine when a junior engineer writes "subpar" docs - this is just normal growth process. Docs that fail to see the "larger picture", where it's apparent that the junior engineers' understanding of the problem statement is wrongfully "how I need to build this", rather than "what is the business problem my feature is addressing, what are the critical aspects of it, what are non-negotiables, what are the constraints I'm working with, and thus why I think these are the areas we need to align on". This is totally okay when a junior engineer does it - the eye for understanding and breaking down problems is something that comes with experience.

But what's not fine is when peers who have 5+ years of tenure is producing the same poor-quality doc. At that point, it's not something solvable with mentorship - it's just pure laziness and lack of care. Poor AI docs have none of that "engineering narrative" and simply is technical jargon after jargon, made-up terminologies, verbose sections, ambiguous sections, code snippets that assumes the reader is already familiar with the code, all simply describing "this is what I will build", and no explanation on how that conclusion was reached.

Conclusion

This post has been more of a rant than anything. And after reading over it a few times, I think my frustration is not AI writing in general, but a frustration specifically towards people who vibes a doc and asks me to "review" their slop and ends up wasting my time.