A colleague asks you a question. You paste it into Copilot, copy the answer, and send it back. Quickly helped? Not really. There's a name for this: a meat proxy.

What is a meat proxy?

A meat proxy is someone who forwards questions to an AI tool and sends the answers back unread. You act as a human intermediary. You add nothing but wait time.

Developer Niklas Gruhn coined the term in his essay "Don't be a meat proxy" from August 3, 2026. The piece spread quickly, including via Simon Willison and Hacker News. Since then, there's even been a dedicated site with a nod: meatproxy.me, featuring a 30-second test to see if you are one.

Why it's not helpful

Gruhn gives two reasons:

  • The other person can do it themselves. Your colleague can also ask a question to an AI tool. It's faster, and then the colleague can determine the context.
  • Reading AI text takes effort. Answers are often long, full of technical jargon, and sometimes wrong in a way that sounds very convincing. If you forward them unread, you pass that task to the recipient.

We add a third: you lose trust. If someone notices that your answers come straight from a chat window, they won't ask you next time.

Recognizable in Microsoft 365

With Copilot in Outlook, Teams, and Word, forwarding is easier than ever. That's the pitfall. Some examples:

  • Responding to a client email with a Copilot draft you haven't reviewed.
  • Answering a question in a Teams channel with a pasted Copilot response, including "Here is an overview of...".
  • Forwarding a meeting summary without checking if the action points are correct.
  • Letting a proposal or advice be written without understanding the content yourself.

Copilot is a powerful tool. But it remains a tool. The value lies in what you do with it.

Here's how not to be a meat proxy

  1. Read the answer. Completely, not just the first line.
  2. Understand it. Can you explain it in your own words? If not, you're not done yet.
  3. Check it. Does it match what you know, the source, the client's situation?
  4. Write your own answer. Concise, in your own words, with your judgment included.
  5. Say what you don't know. An honest "I need to check on that" is worth more than a confident AI answer that's incorrect.

What does this mean for your organization?

A Copilot license doesn't automatically make your team smarter. People must learn how to use AI without turning off their own thinking. That is also a duty: the AI regulation requires employees to be sufficiently AI literate. Read more about AI literacy and the AI regulation.

Start with three steps:

  • Train per role. A sales associate uses Copilot differently than an HR associate. See Copilot training per role.
  • Set agreements. When can an AI answer be sent to a client, and who reads it first? Establish it in your AI governance.
  • Measure usage. Not just how many people use Copilot, but also what it delivers. Read how to measure Copilot adoption.

The right person for the right challenge

At 365 Angels, AI agents are the tools of our Angels, never their replacement. An Angel is always a human: someone who understands the question, checks the answer, and puts their name to it. Exactly the opposite of a meat proxy.

Want your team to use Copilot this way? Contact 365 Angels. We'd be happy to think with you.