Nine Claude Code Subagents Wrote This Blog Post – Can You Tell?

https://benjaminste.in/blog/2025/11/12/everyone-says-they-can-spot-ai-writing/

by benstein • 8 months ago

1 3 8 months ago

burkaman• 8 months ago

Yes, immediately. The em dash and emoji in the title are hints, but I know some humans write like this too.

(what?!)

This is a classic LLM-style exclamation trying to create an emotional reaction where it doesn't really make sense. Something in your prompt told the LLM that this project was supposed to be really surprising and groundbreaking, so now it's going to sprinkle filler language throughout the piece reflecting that, whether or not the specific context is actually surprising.

indefatigable

This adjective does not make sense in this context, but it's the sort of vaguely positive and smart-sounding word that LLMs like to use.

autonmous

This is actually surprising, LLMs don't make typos like this (unless you tell them to). I suspect you did some hand-editing, or this whole post is a prank and you actually did write it all yourself.

the microwave burrito of content creation—technically food, requires minimal effort, and you feel vaguely ashamed afterward.

Hard to explain, but this is the kind of bland but mildly amusing joke that LLMs come up with. I think it's maybe too much explanation? If you have to explain your joke analogy, and there isn't anything particularly unexpected in the explanation, then it wasn't a very good joke.

It’s generic, bland, and sounds like every other piece of AI-generated prose flooding the internet.

True, but then immediately after:

The problem isn’t using LLMs for writing. It’s how we’re using them.

This construction is generic, bland, and sounds like every other piece of AI-generated prose.

I typically have 3 tabs open: Google Docs, my homeboy ChattyG, and El Clauderino (if you’re not into the whole brevity thing).

I don't know if this is an LLM tell, but I definitely rolled my eyes reading it. At this point I would generally stop reading post because I'm not enjoying it and not getting anything useful out of it.

The results are shockingly good. But all that copying, pasting, and reformatting is exhausting. It’s the opposite of what AI tools should be. Like rinsing dishes before putting them in the dishwasher, then hand-washing them again when they come out.

Obvious LLM style, and another uninspired analogy that adds nothing.

Think Ocean’s Eleven, except instead of robbing a casino, they’re stealing back your authentic voice from the abyss of generic AI prose.

You can't tell me you read all these and thought "yeah these are good bits, nobody will know this was an LLM'.

Authenticity Editor hunts AI tells—the distinctive phrases that scream

This is distinctive phrasing that screams LLM.

Plot twist: This entire post was written by the multi-agent system I just described. What you just read wasn’t me writing about the system—it was the system writing about itself.

This is not a plot twist, it's the entire premise that you introduced in the first sentence of the post.

Zero AI tells, removed all em-dash dramatic pauses

There were at least seven em-dash dramatic pauses.

Overall, yes it's obvious this was Claude, and more importantly, the bland filler that Claude added has made your content less interesting and more difficult to get through. I would encourage you to post the hand-written materials that you fed into Claude and see if people like that more or less than the LLM version.

benstein2• 8 months ago

The title was intentionally ironic :-)

benstein2• 8 months ago

I really appreciate how much thought you gave this. Interesting to me that biggest tells came out of the attempts at humor and metaphor.

Nine Claude Code Subagents Wrote This Blog Post – Can You Tell?

Related Stories