alt Hacker NewsDon't be a meat proxy
Comments
The problem here is that the paradigm is AI-assisted development, but a LOT of people are treating it as "AI-independent" development instead, a.k.a "just send it to the agent and blindly trust whatever comes out of it", *including* deferring all responsibility / blame to the AI itself which is absolutely ridiculous.
Well put, this is incredibly becoming annoying. Like has it occured to you i can prompt claude too?
From what I have seen personally over the last 8-12 months, I find this kind of complain ironical in most scenarios. Most of the people complaining this are comfortable with their own usage and distribution of LLM generated content and feel good about it, where as when they hear it from some one else it feels burdensome
If I wanted to know what the AI said, I would have asked it myself.
Don't give me AI feedback at all.
Claude’s tone and style has gotten worse with the recent models, not that it was ever very good. It’s both too concise (stringing together a bunch of jargon/acronyms together in weird ways that aren’t familiar even in areas where you have expertise – with no ramp up, context, or clarity) and too verbose (printing out reams of the word salad mentioned above, often just restating things repeatedly). It feels like writing from someone who really doesn’t know where to start or where to end on a topic so they try to cover their ass by saturation. I suppose that’s not far from reality. Asking it a dozen different ways to speak plainly, use STE, ELI5, whatever has little appreciable effect.
I recently started a new consulting gig and got a 36 page architectural white paper that was authored by Claude. It took me an entire business day and a monstrous headache to decipher and distill it down to a 1 pager. This shit is unsustainable.
I feel like when most LLM type emails from people it’s because of one of three reasons:
1) they don’t know the answer - looked up on llm
2) correlate to #1, I want to look smarter than I actually am
3) I’m too lazY to send you a thoughtful / personal response , I offloaded it to an llm
The problem is a lot of people ask questions that claude et al can objectively answer way better than me, and they didn’t think to ask.
Should I a) give a worse answer b) ask claude and launder it as a meat proxy or c) tell them to ask claude?
b and c piss people off but often get them to a better answer.
A lot of people don't understand what using AI as an output multiplier means. They take it literally, resulting in multiplying busywork across the company.
Misguided HR / people Ops aggravate the situation, by using slack engagement as a productivity measure. Flagging people that are brief but relevant as slackers.
Apparently engineers are running out of tokens with our mediocre token budget, and there is talk of trying to run local LLMs.
Meanwhile execs get unlimited tokens...
And here I am, only automating the most mundane and boring stuff so I can program by hand, which I enjoy.
I got a spreadsheet of requirements for some software from a customer recently.
It was thousands of lines long with repeats and a boatload of conflicting requirements.
When we had a meeting to review it they didn’t know what many of the requirements were. They explained that many people made the spreadsheet but they were all in the meeting and nobody could quite describe some items or how they worked together. The vast majority of the spreadsheet was a mystery.
The phrasing and mishmash of concepts / inability of anyone to explain much of it made me suspect it was largely generated by AI.
Half of this feels like a communication/empathy problem. "Summarise this for someone without our context" fixes the wall of text, and just asking someone to do that might work.
Validation is harder, as nobody skips it because they weren't told to, but because relaying is faster, and unread output looks like validated output right up until it bites. That's probably not solved by asking nicely, I don't think there's a process fix for that beyond people actually being held to it.
My pet peeve is endless Confluence Wiki pages that are clearly written by Claude (or whatever agent they like to use). I get bombarded with these, like it means something. Usually I say: "let me summarize this with my agent", but people rarely get the point.
- how come your blog page doesnt show a list of posts?
- isnt that like standard behavior on any blog?
I read this a couple weeks back and its appeal really stuck with me: if you expect someone to give their attention to something you made, make sure you've given your own attention and effort first.
https://news.ycombinator.com/item?id=48497609 https://tombedor.dev/human-attention-and-human-effort/
Solution Prompt: simply explain to 5 year old kid.
Actually I'm not sure why more people aren't reading and reposting outputs to save people tokens on generating more outputs (I gather that this article is more objecting to the reposting-without-reading)
Expect to see more of this as people's writing skills atrophy. I have done this sometimes when the words aren't coming to me. But, yeah, it'd probably best not not bother with the comment if I can't come up with a coherent articulate comment myself.
Delegating stuff to Google (more recently AI) and repeating back the result, is one of my core functions at work next to programming.
A co-worker I'd just tell to RTFM when it's bothering me. Higher up the hierarchy, the egos are more fragile and I'm afraid the can't take it.
Obviously the solution is to have a SKILL.md which will make an agent translate it to a humanese!
For anyone unfamiliar with the meat reference, here is Terry Bisson's delightful 1991 short-short story, "They're Made out of Meat":
https://www.eastoftheweb.com/short-stories/UBooks/TheyMade.s...
Some version of this needs to be hosted on a dedicated domain like nohello.com that I can respond with as needed. I might just do it
Last week a colleague did this to me; pasted to me in chat "Gemini said". I just ignored it and went with talking about the issue on another thread. I like LLMs and I use them every day, but I'm not gonna answer copy&pasted LLM stuff in human to human conversation. Distill it and let me know your thoughts; I don't need to read you LLM output, I can generate that myself.
Rough opinion: If you're a developer who chose to be a meat proxy, you shouldn't whine if/when you're laid off, because clearly LLMs can do your job.
Lol, obviously never ventured into the Microsoft Techcommunity forums..
I've been on the receiving end of too many "Slop Grenades" as well, to the point where I simply reply with https://noslopgrenade.com/ these days.
While I managed to avoid it for a long time, using generative AI to drive my day to day work life while resisting the pressure to become a meat proxy, I will admit that since joining an "AI native" startup I have become one. Whenever I'm trying to triage 20 different urgent things all day, every day, the economics of personalizing every message just isn't there anymore. It's not realistic. I do still try to way to filter down every AI-generated message into something digestible, but I feel like I'm in a mutually assured destruction type scenario where I can only survive the tidal wave of slop with more slop. I think we're in uncharted territory here in terms of the kind of workload generative AI can drown you in, especially when it's being used without strong discipline.
I do still think there are still places where being a meat proxy is legitimately okay or even desirable, such as PR comments. If everyone's code is generated by AI, and AI will ultimately be the primary consumer of the PR comments, it seems reasonable to me to allow AI to write the comments. Maybe I've just lost my mind though.
I'm surprised this happens between workers within the same knowledge domain. If you're a software dev asking another software dev for an opinion on code, can't you just... talk shop?
In my experience this sort of thing happens all the time when something has to cross between different knowledge domains and the sender doesn't apply their mind in order to co-operate on the issue at hand.
In the pre-LLM past, the sender would add no value, acting as a simple email forwarder, and they would blame you for the delays caused by your inevitable clarificatory questioning. But I always had a defence, which is that your email had no inputs and I had to ask questions to clarify.
Now, though? With LLMs, they just run whatever it is (contract, memo, policy) through whatever LLM they have available and paste the output in an email, giving them the appearance of having done work and added value to the project.
But their LLM outputs don't make sense, or don't apply to our organisation, or is a fluffy and abstract "right answer" with no connection to the specific concerns of the business. Parsing it is a chore, and takes time.
And since now I am the only one actually taking that time, I become the visible cause of the delay.
It's infuriating. Any tips on how to deal with this would be greatly appreciated.
Of course you can relay the output. Dumb to say you can't. In the same way you can relay the output of anything, if it's useful.
But it should form part of your own judgment, not, as you say, just pasted into a chat without any thought.
That is super actual in my team. The biggest problem is that Claude tends to write long prose and use jargon which is not common for us. Often, to understand the whole idea I have to read all the text. The best if the writer rephrases what Claude has written, as wording/jargon will be familiar for the reader and will make communication easier.
I use it to summarize my work and update JIRA/Asana because that is still the thing I hate doing most.
I’ve been saying for a long time that most fights in your personal life are a proxy war but I never thought of a single person as being a meat proxy. Nice.
I append "sound kinda dumb but be factually correct" to my Claude requests and get much less jargony responses.
> frequently contains all too plausible nonsense, and is increasingly jargon dense.
Does anyone know how to deal with the jargon part? It is getting hard to use claude and even worse when someone sends you the direct output from claude.
Why are LLMs producing this super tense text more often now? Is it because they are being optimized to use fewer tokens?
I'll always give people my honest effort and benefit of the doubt initially, but those who violate it are treated likewise. The only way to put down this kind of behavior is to charge it a social cost. If we do not do this, the cost is externalized to everyone else who conducts themselves with care.
> I can talk to Claude myself
OK but can you talk to my specific claude code conversation that has 6 months of context on the mechanism you are struggling with?
The biggest win for AI dev efficiency is cutting down what gets loaded into context. Semantically matching tasks to the top tools helps a lot.
I will frequently post the output of a computer to slack as a part of an engineering discussion. For Claude output I'll treat it the same, wrap it in code blocks, and say something like "Claude's analysis". If its ok to post system logs, why is it less ok to post claude's output (also a computer), especially if I declare this? Is the difference that its plausibly human-level speech and thus breaks an implicit contract?
> "frequently contains all too plausible nonsense"
This really isn’t the case with frontier models in 2026.
I’ve found (sadly) that every time I thought the model was hallucinating, I was in fact the one who was mistaken.
I happen to be currently working on our ISO 27001 implementation and as one of the prerequisites I'm writing our internal AI policy.
I've included a link to this blog post, just for kicks.
Is this the new "let me google that for you"?
Reminds me of this post, from a while back: https://news.ycombinator.com/item?id=48876441
claude said u cant call its users meat proxies. Now back to watching 'ow my balls!'
I am so tired of this. It’s even getting to the point where people without the full context and understanding are meat proxying incorrect information. Reading a bunch of slop that isn’t helpful hinders.
If they instead read it, and distill it down to “have you checked X?” Someone with the full understanding can easily go: “Yes, X doesn’t fit because of this other reason.” within seconds. Sometimes this happens without asking a teammate. However, when I do ask I’m not asking for you to ask Claude and paste the results, I have my own tokens for that. I’m asking you because I think your knowledge will be helpful in finding the answer.
When it comes to cost per implemeting SPEC, agents are far cheaper than human salary. So, it's not going to go well.
That said thankfully there's no way of measuring a good SPEC yet. If there is that will collapse the SWEs profession.
Thank you for saying this. I’ve been dealing with this at work and it’s incredibly frustrating