Niklas Gruhn coins 'meat proxy' for blindly relaying AI output
Niklas Gruhn coined the term meat proxy to describe someone who blindly copies and pastes the output of an AI system and passes it on to another person, acting as a human relay rather than adding anything of their own. Simon Willison picked up the term on his blog (via Lobste.rs) and used it to make a point about how AI tools should be used in communication with other people. His advice: prompting an AI system is fine, but simply forwarding whatever it produces is not. Instead, the person should read the output, understand it, validate it, and then write their own response in their own words. Willison frames that final step, writing the response yourself, as a decent certificate that the first three steps (reading, understanding, validating) actually happened. He presents the extra effort of doing so as the value a person adds on top of what the AI already generated. The post does not name an employer, publication or date for Gruhn, nor does it give a concrete real-world example of a meat proxy incident; it is a short definitional post rather than a reported story.
Key facts
- Niklas Gruhn coined the term "meat proxy" for people who blindly copy and paste AI-generated output to their peers.
- Simon Willison shared the term on his blog, crediting Gruhn and citing Lobste.rs as the source he found it via.
- Willison's advice: prompting an AI system is fine, but relaying its output unread is not.
- He recommends reading, understanding and validating AI output before writing a response in one's own words.
- Writing that response yourself, in his framing, is a "decent certificate" that the prior steps were actually done.
Why it matters
As AI-generated text spreads through emails, chat messages, reports and code reviews, it has become easy for a person to become nothing more than a pass-through for a model's output, forwarding it without engaging with it. Naming that behavior meat proxy gives a compact way to call it out, and Willison's post turns the label into a short practical standard: prompting is fine, but unread relaying is not.
Who it affects
Anyone who uses AI systems to produce text, code or answers that then get handed to another person, whether that is a colleague, a client or a broader audience. The post frames it as a general communication habit rather than something tied to one team, industry or job title.
How to use it
The post offers a simple sequence rather than a tool or product: prompt the AI, then read the output, understand it, validate it, and only then write a response in your own words. There is no pricing, licence or specific workflow attached, just that habit.
How solid is it
The claim is a definitional observation rather than a data-backed finding: Willison names Gruhn as the term's coiner and credits Lobste.rs as where he encountered it, but the post gives no date, no employer or publication for Gruhn, and no documented example of a meat proxy incident. It reads as commentary and advice, not as reported evidence.
Risks and caveats
The post does not say how widespread the behavior is or offer any measurement of it, so the term should be read as a proposed way of talking about a problem rather than a quantified trend. It also does not address cases where forwarding AI output unedited might be appropriate, such as sharing raw data for someone else to evaluate.
“But don't just relay the output. Read it, understand it, validate it, and then write a response in your own words (a decent certificate that you've done the prior steps).”
— Simon Willison