There is a new name for the coworker who responds to a simple question with seven paragraphs that very clearly came straight from ChatGPT: a “meat proxy.”
You probably know the type. They paste a question into an AI chatbot, copy the answer into Slack, and consider their work complete. The response may be long, polished, and packed with suspiciously corporate phrases, but it contains very little evidence that the person sending it has actually read it.
Now (thanks to one software engineer), we finally have a term for that.
What is a meat proxy?
This Tweet is currently unavailable. It might be loading or has been removed.
The term appears to have been coined, or at least popularized, by German software engineer Niklas Gruhn, who laid out the phenomenon in an Aug. 3 blog post.
Gruhn said it kept happening everywhere: He would ask a question in Slack, leave feedback on a code review, or argue with friends in a group chat, and someone would return with, “Claude said: [giant response verbatim].”
The problem was not simply that the answer came from AI; it was that the person sending it had apparently contributed nothing beyond a few copy-and-paste commands. Meanwhile, Gruhn still had to wade through a response that could be verbose, filled with jargon, and confidently wrong.
This Tweet is currently unavailable. It might be loading or has been removed.
As an example, he shared one particularly impenetrable sentence Claude had given him about “NATS control-plane events” and “R3 quorum re-form.” Gruhn said he had to look up nearly every word just to understand what it was trying to say.
His proposed rule was simple: Use AI all you want, but “read it, understand it, and validate it” before responding in your own words. That final step matters less for hiding AI use than for showing an actual human stands behind the answer. Otherwise, Gruhn argued, the person in the middle isn’t really helping. They’re simply serving as a fleshy delivery system — or “meat proxy” — for Claude.
This Tweet is currently unavailable. It might be loading or has been removed.
Gruhn’s post quickly took off on Hacker News, where it attracted more than 700 comments. Developers began sharing their own encounters with coworkers who generate enormous amounts of AI content and leave everyone else to sort through it.
Some described receiving hundreds of lines of AI-generated analysis followed by a request to check whether any of it was correct. Others complained about sprawling internal documents that looked impressive until somebody actually read them. One commenter called these unsolicited blocks of AI text “slop grenades”: Someone tosses a giant pile of chatbot output into the conversation, then walks away while everyone else deals with the mess.
This Tweet is currently unavailable. It might be loading or has been removed.
The problem gets more serious when the AI output is not just an annoying Slack message but actual work someone else has to approve. Before new code is released, another programmer typically reviews it to make sure it works and will not break anything. But if someone submits a pile of AI-generated code they do not understand, the reviewer is suddenly stuck figuring out how it works, catching its mistakes, and essentially finishing the job for them. Software engineering consultant Michaela Greiler has a name for that, too: “code review exploitation.“
Steve Mayne, CTO of tax platform Yonda, described exactly what that can look like in a LinkedIn post about meat proxies: “A pull request lands, it’s plausible, it’s large, but the person who ‘wrote’ it can’t explain line 40.”
This Tweet is currently unavailable. It might be loading or has been removed.
Maybe continues, writing that workplace review processes were designed for a time “when writing was expensive and reading was cheap.” AI has flipped that on its head — producing an impressive amount of work can now take seconds.
That does not mean every sentence beginning with “Chat said…” is a workplace crime. Some people online argue that identifying an answer as AI-generated is useful because it tells the recipient where the information came from. And if someone has read the response, checked it, and decided it is worth sharing, quoting the chatbot is not necessarily the problem.
This Tweet is currently unavailable. It might be loading or has been removed.
At the end of the day, use ChatGPT. Use Claude. Just make sure the answer passes through your brain before it enters the group chat. Otherwise, your coworkers now have the perfect term for you.