The new workplace dynamic nobody's talking about
You ask a question in Slack. A colleague responds. But it's not their words — it's Claude's, pasted verbatim, labeled with a "Claude said:" tagline. Paragraphs of jargon-dense prose that took the sender zero effort to produce and now takes you five minutes to untangle.
This is the meat proxy problem, and it's spreading through every engineering org I can see.
The term comes from a post by Niklas Gruhn that hit the top of Hacker News today (658 points, 291 comments) — the kind of engagement that tells you this is a nerve, not a niche opinion. The meat proxy is someone who slots themselves between an AI and a human conversation, relaying LLM output without adding any value. They're not a collaborator. They're a passthrough.
And the thing is: everyone in the comments has been on the receiving end. Here's what it looks like, why it's worse than you think, and what to do instead.
What a meat proxy actually looks like
The pattern is always the same:
You: Should we use Redis or Memcached for this?
Them (2:16 PM):
Great question! The choice between Redis and Memcached is a nuanced decision that requires careful consideration of multiple factors. Let me break down the key differences: Redis offers a rich set of data structures including strings, hashes, lists, sets, and sorted sets...
Nobody talks like this in Slack. Nobody's grateful for it. As one commenter on the thread put it: "I just ignore people who do that. I don't care what Claude or ChatGPT said. If you cannot be arsed to write yourself then I cannot be arsed to read it."
The worst variant is the code review meat proxy:
Copy/paste the ticket description into Claude Code. Don't look at the code or read what Claude has written. If there's any feedback from reviewers, copy/paste that into Claude Code as well. If necessary, iterate.
That works. But who has done the implementation? The reviewers did, using Claude Code, and you as a meat proxy.
The reviewer becomes the de facto implementer, and the PR author contributes exactly zero understanding of their own diff. The reviewer-vs-author relationship inverts completely.
The "slop grenade" problem
The HN thread spawned its own terminology. One user linked noslopgrenade.com, a single-page site that should be mandatory reading for every team adopting AI tooling:
Stop throwing AI-generated walls of text into conversations.
The site shows a Slack conversation where someone asks a simple question and gets a multi-paragraph AI essay in return. It labels this a "slop grenade" — a verbatim AI dump that explodes into a conversation, wasting everyone's time.
The key insight: AI output is extra effort to read. It's verbose. It's riddled with confidently wrong details. And as Gruhn notes, it's "increasingly jargon dense" — the kind of sentences that force you to look up half the words just to figure out what's being said:
NATS control-plane events: stream leader election / R3 quorum re-form during pod churn.
That sentence is real. It came from Claude. And someone thought sending it to a human was better than writing "our NATS setup had a leadership election issue during the deploy."
Why it's worse than just being annoying
The meat proxy problem has a real cost that goes beyond workplace etiquette. Here's the actual damage:
Cognitive debt. Ankur Sethi wrote a related post yesterday that pairs perfectly with the meat proxy argument. His insight: blindly shipping AI-generated code creates "a colossal amount of cognitive debt." You don't understand your own codebase anymore. You're not building a mental model of the system — you're letting the LLM hold the model and just signing off on its output.
The authority trick. When someone prefaces a response with "Claude said," they're borrowing AI's false authority without doing the work to verify it. As one HN commenter noted: "It's used as a proxy for authority, as in there is some validity in this statement. But it's disguised as humbleness — 'I don't know if this is true, but Claude said xyz, with your expertise you might know more.'" This is worse than a wrong opinion. It's a wrong opinion with a veneer of machine objectivity.
You're training your team to distrust you. Every meat proxy response conditions your colleagues to ignore your messages. They learn, over time, that talking to you means reading AI output they could have generated themselves. You're becoming a filter that degrades signal quality.
The fix is simple and hard
The rule has exactly one sentence: Read it, understand it, validate it, then write a response in your own words.
That's it. There's no trick. Use AI as much as you want — for research, for drafting, for first passes. Just don't stop before the last step. Writing the response yourself is "a decent certificate that you've done the prior steps," as Gruhn puts it. It proves you understood what the AI told you.
The HN thread surfaced a good heuristic from user Versipelle: "If you feel the need to specify in your answer that it is AI generated, work it more. Nobody cares whether you used your memory and expertise alone, asked a peer, googled it, or asked ChatGPT to refine your answer."
Here's a practical workflow that respects both speed and quality:
# Don't do this:
def respond_to_coworker(question):
llm_response = ask_claude(question)
slack.send(f"Claude said: {llm_response}") # meat proxy mode
# Do this instead:
def respond_to_coworker(question):
llm_response = ask_claude(question)
understanding = synthesize(llm_response, context=question)
your_words = distill(understanding) # short, in your voice
slack.send(your_words)
The second version takes a few extra minutes but preserves your reputation, your team's trust, and your own understanding of the system you're building.
The deeper problem
The meat proxy phenomenon isn't a technology failure. It's a social contract failure between people who owe each other attention.
One commenter put it better than I can: "If you expect someone to give their attention to something you made, make sure you've given your own attention and effort first."
The tools aren't going to get worse at generating text. They're going to get better. The "make sure you've done the work" norm isn't optional — it's the only thing that keeps AI from degrading the quality of human communication in every channel it touches.
The meat proxy is a choice. Don't be one. And if someone sends you a slop grenade, you can do what one HN commenter does: "Thanks, but I can ask Claude myself."
Sources:
Top comments (0)